Influence calculation method and influence calculation program

Abstract

The ability to specifically demonstrate the influence of a particular factor on a target factor on a directed cyclic graph. [Solution] The influence calculation device 100 extracts a circulating path P from a directed circulating graph G that includes the target factor T and the focus factor t, and classifies the circulating path P into self-reinforcing circulating paths where the product of the causal coefficients is positive, and balanced circulating paths where the product is negative. The strength of the self-reinforcing circulating path is then calculated as the strength of the focus factor t's circulating influence E on the target factor T in the promotion direction. The strength of the balanced circulating path is then calculated as the strength of the focus factor t's circulating influence E on the target factor T in the suppression direction.

Description

[Technical Field]

[0001] This invention relates to a method for calculating influence and a program for calculating influence. [Background technology]

[0002] In recent years, various types of data have become available, and data analysis has become crucial for improving issues and achieving results by making appropriate data-driven decisions. Techniques for analyzing multi-item data and extracting causal relationships between factors include: 1. A directed acyclic graph that represents the relationships between factors assuming no cycles (loops), or 2. A directed cyclic graph that represents the relationships between factors while allowing cycles.

[0003] Prior art includes, for example, a directed graph defined by nodes representing problems and links representing causal relationships between nodes, which also maintains a common activity goal, and displays loops and links containing nodes that represent the common activity goal among the nodes included in the directed graph. There is also a system that extracts one of several loops as a point requiring resolution based on the attributes of nodes, links, and loops, highlights the extracted loop, and displays countermeasures for resolution based on the attributes of the extracted loop, node, and link. Furthermore, there is a system that displays causal nodes or resulting nodes in different display modes depending on the degree of sensitivity that causal nodes have to resulting nodes (see, for example, Patent Documents 1 to 3 below). [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2024-49971 [Patent Document 2] International Publication No. 2016 / 151828 [Patent Document 3] Japanese Patent Publication No. 2016-224548 [Overview of the project] [Problems that the invention aims to solve]

[0005] While directed cyclic graphs can represent complex relationships that can occur in reality by allowing cycles and expressing the relationships between factors, it is difficult for users to interpret (gain insight into) the complex relationships between many factors. Conventional technologies cannot clearly show the influence of a particular factor on a target factor (objective factor) in a cycle path that includes the target factor and each other factor on a complex directed cyclic graph. Therefore, there is a need for concrete support, such as effective presentation, to appropriately interpret directed cyclic graphs.

[0006] In one aspect, the present invention aims to specifically demonstrate the influence of a particular factor on a target factor on a directed cyclic graph. [Means for solving the problem]

[0007] According to one embodiment, the influence calculation method and influence calculation program are required to include the following processes: extracting a circulating path from a directed circulating graph that includes a target objective factor and a focus factor of interest; classifying the circulating path into a self-reinforcing circulating path where the product of the causal coefficients is positive and a balanced circulating path where the product is negative; and calculating the strength of the self-reinforcing circulating path as the strength of the focus factor's circulating influence on the target factor in the promoting direction. [Effects of the Invention]

[0008] According to one embodiment, it becomes possible to specifically present the influence of a factor of interest on a target factor on a directed cyclic graph. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is an explanatory diagram of one embodiment of the influence calculation according to the embodiment. [Figure 2]Figure 2 is a diagram showing a directed acyclic graph and a directed cyclic graph. [Figure 3] Figure 3 is an explanatory diagram for defining the characteristics of a cyclic path. [Figure 4] Figure 4 is an explanatory diagram of cyclic influence power. [Figure 5] Figure 5 is an explanatory diagram of a specific example using cyclic influence power. [Figure 6] Figure 6 is a block diagram showing an example of the function of influence power calculation according to the embodiment. [Figure 7] Figure 7 is a block diagram showing an example of the hardware configuration of the influence power calculation device. [Figure 8] Figure 8 is a flowchart showing a processing example of the influence power calculation device. [Figure 9] Figure 9 is an explanatory diagram showing the strength of cyclic influence power. [Figure 10] Figure 10 is a chart showing an example of analysis target data. [Figure 11] Figure 11 is a chart showing an example of a directed cyclic graph. [Figure 12] Figure 12 is a chart showing another example of a directed cyclic graph. [Figure 13] Figure 13 is a diagram showing an output example of the influence power calculation device. [Mode for Carrying Out the Invention]

[0010] Hereinafter, embodiments of the influence power calculation method and the influence power calculation program according to the present invention will be described in detail with reference to the drawings. In the embodiment, cyclic paths including a target factor (target factor) and each focused factor (focused factor) are extracted from a directed cyclic graph, and these cyclic paths are classified into a self-reinforcing type and a balanced type. For a self-reinforcing type cyclic path, the strength is calculated as the strength in the promoting direction of the cyclic influence power of the focused factor. For a balanced type cyclic path, the strength is calculated as the strength in the suppressing direction of the cyclic influence power of the focused factor.

[0011] In this embodiment, the strength of the calculated cyclic influence makes it possible to clearly indicate whether the influence of a target factor on the entire directed cyclic graph (spiral) is strengthened or weakened by the causal cycle through each factor of interest, from a complex directed cyclic graph. For example, by calculating the cyclic influence on the entire directed cyclic graph, the influence of the factor of interest on the target factor can be specifically shown. In this invention, for example, an influence calculation device 100 such as a general-purpose PC or server performs the processing related to influence calculation.

[0012] (Embodiment: One embodiment of the method for calculating influence) Figure 1 is an explanatory diagram of one embodiment of influence calculation according to the embodiment. The influence calculation device 100 generally performs the following processes (1) to (4) in sequence (details of each process will be described later).

[0013] (1) Obtain the data to be analyzed from a data series consisting of multiple factors (items), perform causal analysis on the data to be analyzed, and create a directed cyclic graph G (see Figure 1(a)).

[0014] (2) Extract a travel path P that includes the target factor T and the focus factor t. The target factor T is the factor used to calculate the influence, and in the example in Figure 1(a), it is assumed to be "operating profit".

[0015] The focus factor t is one of several factors that are considered for causal analysis with the target factor T on the travel path P.

[0016] A cyclical path P is a single loop (path) on a directed cyclical graph G that starts at the objective factor "operating profit" T, passes through multiple factors, and returns to the objective factor T "operating profit" again. In the example in Figure 1, multiple factors such as "employee turnover rate," "sense of fulfillment," and "overtime hours" can each be the focus factors t.

[0017] (3) Classify the circulating paths P into self-reinforcing and balanced types. The influence calculation device 100 extracts circulating paths P from the directed circulating graph G that include both the objective factor T and the focus factor t, and classifies them into self-reinforcing and balanced types based on the sign of the product of their causal coefficients.

[0018] For example, if one circulating path P1 extracted on a directed circulating graph G has the objective factor T "operating profit" → the focus factor t "employee turnover rate", then we extract circulating paths P that include both the objective factor T "operating profit" and the focus factor t "employee turnover rate". In the example in Figure 1(a), it is assumed that three circulating paths P1 to P3 (the loop in P3 is not shown) are extracted that include both the objective factor T "operating profit" and the focus factor t "employee turnover rate".

[0019] The influence calculation device 100 then calculates the product of the causal coefficients of the patrol path P1 (objective factor T "operating profit" → focus factor t "sense of fulfillment" → t "turnover rate" → T "operating profit"), and if the calculation result is positive, it classifies this patrol path P1 as self-reinforcing.

[0020] Furthermore, the product of the causal coefficients of the patrol path P2 (objective factor T "operating profit" → focus factor t "overtime hours" → t "employee turnover rate" → T "operating profit") is calculated, and if the calculation result is negative, this patrol path P2 is classified as a balanced type.

[0021] Furthermore, for the patrol path P3 (objective factor T "operating profit" → focus factor t "overtime hours" → t "sense of fulfillment" → t "turnover rate" → T "operating profit"), if the product of the causal coefficients is negative, this patrol path P3 is classified as a balanced type.

[0022] In the example in Figure 1, each cycle path P extracted when the focus factor t is the factor "sense of fulfillment," and each cycle path P extracted when the focus factor t is the factor "overtime hours," are also classified into self-reinforcing and balanced types, respectively.

[0023] (4) The quantity and quality of self-reinforcing and balanced circulation paths P are compared, and the strength of the self-reinforcing circulation path is calculated as the strength of the circulation influence of the factor of interest.

[0024] The influence calculation device 100 calculates the circulating influence based on the number of self-reinforcing and balanced paths that include each focus factor t, when the objective factor T is "operating profit" (see Figure 1(b)). For example, the circulating influence of "focus factor" t is calculated based on: circulating influence = number of self-reinforcing paths / total number of paths.

[0025] for example, The influence of the focus factor t "employee turnover rate" on the cycle = number of self-reinforcing paths 1 / total number of paths 3 = 33% The influence of the focus factor t "sense of fulfillment" on the circulation of paths = number of self-reinforcing paths 1 / total number of paths 2 = 50% The influence of the focus factor t "overtime hours" on the cycle is calculated as follows: Number of self-reinforcing paths 0 / Total number of paths 2 = 0%.

[0026] Then, as shown in Figure 1(b), the influence calculation device 100 displays a graph (chart) to the user, for example, with the horizontal axis representing each focus factor t and the horizontal axis representing the strength of the circulating influence E.

[0027] The direct causal relationship was analyzed to be that operating profit increases as the turnover rate decreases. However, the above calculation results by the influence calculation device 100 showed that the patrol influence of t "turnover rate" is weak (there are many balanced patrol paths), indicating that simply taking measures to reduce the turnover rate is not effective, and that more appropriate measures can be implemented.

[0028] In the explanation of the process in (4) above, we described the process for the strength of self-reinforcing circulating paths after classification. However, for balanced circulating paths after classification, the strength is calculated as the strength of the factor of focus in the direction of suppressing the circulating influence.

[0029] The influence calculation method according to this embodiment can present the cyclic influence across the entire directed cyclic graph (spiral), showing the influence of factors on the target factor. Furthermore, it can specifically present users with information that allows them to easily identify factors of interest in data analysis (information for interpreting the directed cyclic graph).

[0030] (Regarding the background technology) Here, we will explain the background technology that forms the basis of this embodiment.

[0031] (Regarding directed acyclic graphs and directed cyclic graphs) Figure 2 shows directed acyclic graphs and directed cyclic graphs. Figure 2(a) shows a directed acyclic graph Gx, assuming that there are no cycles (loops) in the causal relationships of the factors. In a directed acyclic graph Gx, the causal graph is relatively simple and easy to interpret.

[0032] Figure 2(b) shows a directed cyclic graph G. By allowing cycles (loops) in the causal relationships of factors, it is possible to represent complex relationships that can occur in reality, but this makes it difficult to interpret. In this embodiment, processing is performed on the directed cyclic graph G shown in Figure 2(b), in which cycles (loops) are allowed in the causal relationships of factors. In the following explanation, the thick lines between factors in each directed cyclic graph G indicate a causal coefficient of - (negative).

[0033] (Travel Paths: Self-Reinforcing and Balanced Types) Figure 3 is an explanatory diagram of the definition of the characteristics of a cyclic path. Regarding cyclic paths, we will explain the self-reinforcing type shown in Figure 3(a) and the balanced type shown in Figure 3(b). In a cyclic path P that forms a single loop on a directed cyclic graph G, the type of cyclic path is defined by the following formula.

[0034] We define a self-reinforcing type as one where the product of the causal coefficients = (-e) × d × (-c) × b × a is positive. A balanced type is defined as a case where the product of the causal coefficients = (-f) × (-g) × h × (-i) × a is negative.

[0035] In a self-reinforcing cyclical pathway, an increase (or decrease) in factor (target factor) T creates a causal relationship that further increases (or decreases) factor T, thus reinforcing the strength of factor T.

[0036] On the other hand, in a balanced cyclical pathway, an increase (or decrease) in factor T leads to a causal relationship that indirectly reduces (or increases) factor T, thus suppressing the strength of factor T.

[0037] (Regarding the influence of the patrol) Figure 4 is an explanatory diagram of circulating influence. In this embodiment, "circulating influence" is newly defined. The difference between this circulating influence and the direction and strength of causal relationships will be explained. In Figure 4, the left half of the directed circulating graph shows a self-reinforcing circulating path P1, and the right half shows a balanced circulating path P2.

[0038] Suppose that, as a direct causal relationship, an increase in the strength of factor t (the factor of interest t) leads to an increase in the strength of factor T (corresponding to the objective factor T), and this is observed to be reflected in a large coefficient. However, when considering cyclical causal relationships from a medium- to long-term perspective,

[0039] 1. If the cyclical path P1 is self-reinforcing, the influence is strengthened through the circulation via factor t = the cyclical influence becomes large. If the causal coefficient of factor t becomes positive in a self-reinforcing type, factor T can be made positive, and in a roundabout way, factor t becomes positive again, and factor T becomes positive.

[0040] 2. Conversely, if the circulating path P2 is balanced, the influence weakens as it passes through factor t = the circulating influence becomes small. In a balanced path, if the causal coefficient of factor t is made positive, factor T can be made positive, but as it goes through the cycle, factor t becomes negative, and factor T becomes negative.

[0041] Therefore, the existence of various cyclical paths P indicates the magnitude of the influence that factor T receives from the cyclical causal relationships that circulate through each factor t, separate from the strength of the causal relationship (symbol a in the diagram).

[0042] As described above, in this embodiment, by using the concept of cyclic influence, it becomes possible to hypothesize, independently of causality, that if the cyclic influence is strong, for example, that factor (focus factor) t is a major factor influencing factor (target factor) T. Furthermore, it becomes possible to derive new interpretations, such as the need for measures focusing on factor (focus factor t) in order to improve factor (target factor T).

[0043] (Specific examples using circulating influence) Figure 5 is an explanatory diagram illustrating a specific example using patrol influence. Similar to Figure 4, the left half of the directed patrol graph shows a self-reinforcing patrol path P1, and the right half shows a balanced patrol path P2. As a direct causal relationship, it was analyzed that operating profit increases as the turnover rate decreases. The relationship between the strength of patrol influence and the patrol path will be explained.

[0044] In the example shown in Figure 5, in the self-reinforcing rotational path P1, a negative turnover rate results in a positive operating profit, which in turn leads to a further negative turnover rate and a positive operating profit. On the other hand, in the balanced rotational path P2, a negative turnover rate results in a positive operating profit, which in turn leads to a positive turnover rate and a negative operating profit.

[0045] Analysis result 1) There are many self-reinforcing circulating paths P1 = a situation where circulating influence is strong. For example, the improvement in operating profit suggests that employee satisfaction is increasing and the turnover rate is decreasing. Therefore, it can be interpreted that continuing measures to reduce the turnover rate is an effective way to improve operating profit.

[0046] Analysis result 2) There are many balanced patrol paths P2 = a situation where patrol influence is weak. For example, this suggests that an increase in operating profit is leading to an increase in overtime hours and a rise in employee turnover. This can be interpreted as indicating that simply reducing employee turnover is not enough to improve operating profit; improvements involving other factors are also necessary.

[0047] (Example of an influence calculation function) Figure 6 is a block diagram showing an example of the influence calculation function according to the embodiment. The influence calculation device 100 includes the functions of a causal relationship calculation unit 601, a recording unit 602, and a circulating influence calculation unit 603.

[0048] The causal relationship calculation unit 601 acquires the data to be analyzed D, calculates the causal relationships between factors, and stores the directed cyclic graph (causal graph) G in the recording unit 602 such as a database (DB). The causal relationship calculation unit 601 performs the process of creating the directed cyclic graph G described above (1).

[0049] The circulating influence calculation unit 603 obtains the objective factor T and calculates the circulating influence E based on the created directed circulating graph G. The circulating influence calculation unit 603 performs the following processes: (2) extracting circulating paths P that include the objective factor T and the focus factor t, (3) classifying the circulating paths P into self-reinforcing and balanced types, and (4) calculating the strength of the self-reinforcing and balanced circulating paths P as the strength of the circulating influence E of the focus factor t. The circulating influence calculation unit 603 outputs the calculated strength of the circulating influence E to the user, for example, by displaying a graph (chart) as shown in Figure 1(b).

[0050] (Example hardware configuration for an influence calculation device) Figure 7 is a block diagram showing an example of the hardware configuration of an influence calculation device. In Figure 7, the influence calculation device 100 can use a general-purpose computer, such as a server or PC. The influence calculation device 100 includes a CPU (Central Processing Unit) 701 such as a processor, memory 702, network IF 703, recording medium IF 704, and recording medium 705. Each component is connected by a bus 700.

[0051] Here, the CPU 701 is a control unit that oversees the overall control of the influence calculation device 100. The CPU 701 may have multiple cores. The memory 702 includes, for example, ROM (Read Only Memory), RAM (Random Access Memory), and flash ROM. Specifically, for example, the flash ROM stores the OS program, the ROM stores the application program, and the RAM is used as the work area for the CPU 701. The program stored in the memory 702 is loaded into the CPU 701, causing the CPU 701 to execute the coded process.

[0052] The Network IF703 is connected to a network NW via a communication line and can connect to an external computer via the network NW. The Network IF703 manages the interface between the network NW and the device's internal components, controlling the input and output of data from the external computer. The Network IF703 can utilize, for example, a modem or LAN adapter.

[0053] The recording medium IF704 controls the reading / writing of data to the recording medium 705 according to the control of the CPU 701. The recording medium 705 stores the written data. Examples of recording media 705 include magnetic disks such as HDDs, optical disks, SSDs (Solid State Drives), and USB (Universal Serial Bus) memory.

[0054] In addition to the components described above, the influence calculation device 100 may also have, for example, an input device, a display, etc.

[0055] The functions of the causal relationship calculation unit 601 and the cyclic influence calculation unit 603 shown in Figure 6 can be realized by programming the CPU 701 shown in Figure 7. The data to be recorded in the recording unit 602 shown in Figure 6 can be recorded using the memory 702 and recording medium 705 shown in Figure 7. The functions related to data acquisition by the causal relationship calculation unit 601 and data output by the cyclic influence calculation unit 603 shown in Figure 6 can be realized, for example, using the network IF 703 shown in Figure 7.

[0056] (Example of processing in the embodiment) Next, an example of the processing performed by the influence calculation device 100 according to the embodiment will be described.

[0057] Figure 8 is a flowchart showing an example of the processing performed by the influence calculation device. The processing shown in Figure 8 is performed by each function of the influence calculation device 100 shown in Figure 6 (CPU 701 in Figure 7).

[0058] First, the influence calculation device 100 performs causal analysis on a data series consisting of multiple items (factors) and calculates a directed cyclic graph G (step S801).

[0059] Next, the influence calculation device 100 extracts a travel path P that includes the target factor (target factor T) and the factor of interest (focus factor t) (step S802).

[0060] Next, the influence calculation device 100 classifies the extracted patrol paths P into self-reinforcing patrol paths P1 and balanced patrol paths P2 (step S803).

[0061] The influence calculation device 100 then compares the quantity and quality of the self-reinforcing circulating path P1 and the balanced circulating path P2, and calculates the strength of the self-reinforcing circulating path P1 as the strength of the circulating influence E of the factor of interest (factor of interest t) in the promotion direction (step S804). The influence calculation device 100 then completes the above process.

[0062] Furthermore, although not shown in the diagram for step S804, the influence calculation device 100 also calculates the strength of the balanced circulating path P2 as the strength of the circulating influence E of the factor of interest (factor of interest t) in the suppressive direction.

[0063] Figure 9 is an explanatory diagram illustrating the strength of the circulating influence. As shown in Figure 9(a), the circulating path P, which includes the objective factor T and the focus factor t, includes a self-reinforcing circulating path P1 and a balanced circulating path P2. The number of circulating paths in the self-reinforcing circulating path P1 (solid line) and the number of circulating paths in the balanced circulating path P2 (dotted line) are shown.

[0064] As shown in Figure 9(b), the strength of the circulating influence E of factor (focus factor) t on factor (objective factor) T takes a value in the range of 0 to 100. The closer the value is to 100, the better the quality (or quantity) of the self-reinforcing circulating path P1. The closer the value is to 0, the better the quality (or quantity) of the balanced circulating path P2.

[0065] The strength of the patrol path in the process performed by the influence calculation device 100 (step S804 in Figure 8) can be calculated, for example, by any of the following I1 (equation (1)) to I4 (equation (4)), or a combination thereof.

[0066]

number

[0067] Regarding the quantity and quality of the patrol paths, for example, I1 and I3, which depend on the number of patrol paths, represent the quantity of patrol paths. On the other hand, I2 and I4, which depend on the causal coefficient and the average number of nodes, represent the quality of the patrol paths.

[0068] (Specific example of processing 1) Below, a specific example of processing using the influence calculation device of the embodiment will be described in the order shown in Figure 6.

[0069] FIG. 10 is a diagram showing an example of analysis target data. The analysis target data D (see FIG. 6) acquired by the influence calculation device 100 includes respective values of operating profit, turnover rate, sense of fulfillment, and overtime hours for each data name (quarterly data name of each department).

[0070] The influence calculation device 100 (causality calculation unit 601) executes the process of creating (1) the directed cyclic graph G described above (step S801 in FIG. 8). The influence calculation device 100 sets the following assumptions and performs machine learning on the coefficient a ij .

[0071] X1 (operating profit) = a 12 X2 + a 13 X3 + a 14 X4 X2 (turnover rate) = a 21 X1 + a 23 X3 + a 24 X4 X3 (sense of fulfillment) = a 31 X1 + a 32 X2 + a 34 X4 X4 (overtime hours) = a 41 X1 + a 42 X2 + a 43 X3

[0072] FIG. 11 is a diagram showing an example of a directed cyclic graph. The coefficients a calculated above ij form a directed cyclic graph G represented by the adjacency matrix shown in FIG. 11(a). Further, FIG. 11(b) shows the directed cyclic graph G represented as a graph. Elements other than the coefficient 0 in the adjacency matrix of FIG. 11(a) correspond to the arrows in FIG. 11(b).

[0073] Next, the influence calculation device 100 (cyclic influence calculation unit 603) performs the process of (2) extracting the cyclic path P including the target factor T and the focused factor t (step S802 in FIG. 8), and the process of (3) classifying the cyclic path P into a self-reinforcing type and a balanced type (step S803 in FIG. 8).

[0074] First, assuming the target factor T is "operating profit" and the factor t of interest is "employee turnover rate," we extract paths containing both from the directed cyclic graph G and classify them into self-reinforcing and balanced types based on the sign of the product of their causal coefficients, as shown below. 11) Operating profit → Employee satisfaction → Turnover rate → Operating profit (Self-reinforcement type) 12) Operating profit → Overtime hours → Employee turnover rate → Operating profit (balanced type) 13) Classify into: operating profit → overtime hours → sense of fulfillment → turnover rate → operating profit (balanced type).

[0075] Furthermore, if we define the factor of interest (focus factor) t as "sense of fulfillment," 21) Operating profit → sense of fulfillment → turnover rate → operating profit (self-reinforcing type) 22) Classify into: operating profit → overtime hours → sense of fulfillment → turnover rate → operating profit (balanced type).

[0076] If we define the factor of interest (focus factor) t as "overtime hours", 31) Operating profit → Overtime hours → Employee turnover rate → Operating profit (balanced type) 32) Classify into: operating profit → overtime hours → sense of fulfillment → turnover rate → operating profit (balanced type).

[0077] Next, the influence calculation device 100 (circulating influence calculation unit 603) performs the process of calculating the strength of the self-reinforcing and balanced circulating paths P as the strength of the circulating influence E of the focus factor t (step S804 in Figure 8).

[0078] The target factor T is defined as "operating profit," and the circulating influence E is calculated based on the number of self-reinforcing and balancing paths that include each factor t of interest. The influence of employee turnover on the cycle E1 = number of self-reinforcing paths 1 / total number of paths 3 = 33% The circulating influence of a sense of fulfillment, E2, is equal to the number of self-reinforcing paths (1) / the total number of paths (2) = 50%. The influence of overtime hours on patrols, E3, is calculated as follows: Number of self-reinforcing paths: 0 / Total number of paths: 2 = 0%

[0079] While the direct causal relationship is analyzed as a decrease in employee turnover leading to increased operating profit, the above analysis reveals that the turnover rate's influence on employee turnover (E1) is weak (indicating a balanced turnover path), suggesting that simply reducing the turnover rate is not an effective strategy. For example, this suggests that increased operating profit is leading to increased overtime hours, which in turn increases the turnover rate. This leads to the conclusion that reducing employee turnover alone is not effective in improving operating profit, and that improvements involving other factors are necessary.

[0080] (Example of processing 2) Figure 12 is a diagram showing another example of a directed cyclic graph. Figure 12(a) is similar to Specific Example 1 (Figure 11), but some of the causal relationships are different, a 34 The case where is positive is shown, and Figure 12(b) shows the directed cyclic graph G as a graph.

[0081] The influence calculation device 100 (circulating influence calculation unit 603) performs the following processes: (2) extracting a circulating path P that includes the objective factor T and the focus factor t (step S802 in Figure 8), and (3) classifying the circulating path P into self-reinforcing and balanced types (step S803 in Figure 8).

[0082] If the target factor T is "operating profit" and the factor t we are focusing on is "employee turnover rate", 11) Operating profit → Employee satisfaction → Turnover rate → Operating profit (Self-reinforcement type) 12) Operating profit → Overtime hours → Employee turnover rate → Operating profit (balanced type) 13) Classify into: operating profit → overtime hours → sense of fulfillment → turnover rate → operating profit (self-reinforcement type).

[0083] If we define the factor we are focusing on (focus factor) t as "sense of fulfillment," 21) Operating profit → sense of fulfillment → turnover rate → operating profit (self-reinforcing type) 22) Classify into: operating profit → overtime hours → sense of fulfillment → turnover rate → operating profit (self-reinforcement type).

[0084] If we define the factor of interest (focus factor) t as "overtime hours", 31) Operating profit → Overtime hours → Employee turnover rate → Operating profit (balanced type) 32) Classify into: operating profit → overtime hours → sense of fulfillment → turnover rate → operating profit (self-reinforcement type).

[0085] Next, the influence calculation device 100 (circulating influence calculation unit 603) performs the process of calculating the strength of the self-reinforcing and balanced circulating paths P as the strength of the circulating influence E of the focus factor t (step S804 in Figure 8).

[0086] The target factor T is defined as "operating profit," and the circulating influence E is calculated based on the number of self-reinforcing and balancing paths that include each factor t of interest. The influence of employee turnover on the cycle E1 = number of self-reinforcing paths 2 / total number of paths 3 = 66% The circulating influence of a sense of fulfillment, E2, is equal to the number of self-reinforcing paths (2) / the total number of paths (2) = 100%. The influence of overtime hours on the patrol is E3 = number of self-reinforcing paths 0 / total number of paths 2 = 50%.

[0087] Here, the direct causal relationship is analyzed to be that operating profit increases as the turnover rate decreases, and the turnover rate also has a strong influence on employee turnover (many self-reinforcing turnover paths). Based on the above analysis, it is suggested that measures to reduce the turnover rate and measures to increase operating profit continue to have a positive impact on each other, thus indicating an effective situation. For example, it can be suggested that the improvement in operating profit increases employee satisfaction and reduces the turnover rate, and it can be concluded that continuing measures to reduce the turnover rate is effective in improving operating profit.

[0088] Figure 13 shows an example of the output of the influence calculation device. Figure 13(a) is a graph output by the influence calculation device 100, which outputs the circulating influence E on the influence benefit of the target factor T. The vertical axis of the graph displays the circulating influence E for each factor of interest (focus factor) t as a bar from 0 to 100%. Figure 13(b) shows the directed circulating graph G output in correspondence with Figure 13(a).

[0089] Apart from the causal relationships represented by the directed cyclic graph G in Figure 13(b), the cyclic influence graph E shown in Figure 13(a) suggests that, for example, for "operating profit," factors with large values ​​such as "NPS" and "sense of fulfillment" can be interpreted as factors with a large influence (factors that amplify each other's influence) in the entire causal loop (spiral). On the other hand, factors with small values ​​such as "turnover rate" and "costs" can be interpreted as factors with a small influence (factors that suppress each other's influence) in the entire causal loop.

[0090] The influence calculation method according to the embodiment described above is applicable to various data analysis techniques that analyze multi-item data and extract causal relationships between factors, and can specifically show the influence of the target factor on the entire directed cyclic graph (spiral) through the causal cycle of each factor of interest.

[0091] As described above, the embodiment extracts a circulating path from a directed circulating graph that includes the target objective factor and the focus factor, classifies the circulating paths into self-reinforcing circulating paths where the product of the causal coefficients is positive and balanced circulating paths where the product is negative, and calculates the strength of the self-reinforcing circulating path as the strength of the focus factor's circulating influence on the objective factor in the promotion direction. By calculating the circulating influence of the entire directed circulating graph in this way, it becomes possible to concretely present the influence of the focus factor on the objective factor. This makes it possible, for example, to easily identify the focus factor.

[0092] Furthermore, the calculation process determines the strength of a balanced travel path as the strength of the focus factor's influence on the target factor in the direction of suppression. By calculating different travel influences for each travel path classification result in this way, it becomes easier to identify the focus factor for each travel path classification result.

[0093] Furthermore, the calculation method involves determining the strength of the circulating influence as the ratio of the strength of self-reinforcing circulating paths to the strength of balanced circulating paths. For example, the calculation method could be based on the product of the causal coefficients of the circulating paths, on the number of self-reinforcing and balanced circulating paths, on the number of circulating paths containing the objective factor, or on the average number of nodes in the circulating paths. In this way, the strength of the circulating influence can be easily obtained by the ratio of strengths for each classified circulating path.

[0094] The influence calculation method described in this embodiment can be implemented by executing a pre-prepared program on a computer such as a PC or workstation. The influence calculation program described in this embodiment is recorded on a computer-readable recording medium and executed by reading it from the recording medium by the computer. Furthermore, the influence calculation program described in this embodiment may be distributed via a network such as the Internet.

[0095] With regard to the embodiments described above, the following additional information is disclosed.

[0096] (Note 1) Extract the cyclic path from the directed cyclic graph that includes the target objective factor and the focus factor, The aforementioned travel paths are classified into self-reinforcing travel paths where the product of causal coefficients is positive, and balanced travel paths where the product is negative. The strength of the self-reinforcing circulation path is calculated as the strength of the circulation influence of the focus factor on the objective factor in the direction of promotion. A method for calculating influence, characterized in that the processing is performed by a computer.

[0097] (Note 2) The calculation process described above is: The strength of the balanced circulation path is calculated as the strength of the focus factor's influence on the objective factor in the direction of suppression. The method for calculating influence as described in Appendix 1, characterized by including processing.

[0098] (Appendix 3) The calculation process described above is: The strength of the self-reinforcing circulating path and the strength of the balanced circulating path are calculated as the strength of the circulating influence. The method for calculating influence as described in Appendix 2, characterized by including processing.

[0099] (Note 4) The calculation process described above is: Calculated based on the product of the causal coefficients of the aforementioned circuit paths, The method for calculating influence as described in Appendix 3, characterized by the features described herein.

[0100] (Note 5) The calculation process described above is: The calculation is based on the number of self-reinforcing patrol paths and the number of balanced patrol paths. The method for calculating influence as described in Appendix 3, characterized by the features described herein.

[0101] (Appendix 6) The calculation process described above is: Calculated based on the number of pathways including the aforementioned objective factor, The method for calculating influence as described in Appendix 3, characterized by the features described herein.

[0102] (Appendix 7) The calculation process described above is: Calculated based on the average number of nodes in the aforementioned circuit path, The method for calculating influence as described in Appendix 3, characterized by the features described herein.

[0103] (Note 8) Obtain data series for each item corresponding to multiple factors, A process to create the directed cyclic graph based on the causal coefficients between each factor in the acquired data series. The method for calculating influence as described in Appendix 1, characterized by including the following:

[0104] (Note 9) Extract the cyclic path from the directed cyclic graph that includes the target objective factor and the focus factor, The aforementioned travel paths are classified into self-reinforcing travel paths where the product of causal coefficients is positive, and balanced travel paths where the product is negative. The strength of the self-reinforcing circulation path is calculated as the strength of the circulation influence of the focus factor on the objective factor in the direction of promotion. An influence calculation program characterized by having a computer perform the processing. [Explanation of Symbols]

[0105] 100 Influence calculation device 601 Causal Relationship Calculation Unit 602 Records Department 603 Traveling influence calculation unit 701 CPU 702 memory 703 Network Interface 705 Recording media D Data to be analyzed E Traveling influence G directed cyclic graph P Loop Pass P1 Self-reinforcing circulating path P2 Balanced Circulation Path T objective factor t Focus Factor

Claims

1. From a directed cyclic graph, extract cyclic paths that include the target objective factor and the focus factor. The aforementioned travel paths are classified into self-reinforcing travel paths where the product of causal coefficients is positive, and balanced travel paths where the product is negative. The strength of the self-reinforcing circulation path is calculated as the strength of the circulation influence of the focus factor on the objective factor in the direction of promotion. A method for calculating influence, characterized in that the processing is performed by a computer.

2. The calculation process described above is: The strength of the balanced circulation path is calculated as the strength of the focus factor's influence on the objective factor in the direction of suppression. The method for calculating influence according to claim 1, characterized by including processing.

3. The calculation process described above is: The strength of the self-reinforcing circulating path and the strength of the balanced circulating path are calculated as the strength of the circulating influence. The method for calculating influence according to claim 2, characterized by including processing.

4. The calculation process described above is: Calculated based on the product of the causal coefficients of the aforementioned circuit paths, The method for calculating influence according to feature 3.

5. The calculation process described above is: The calculation is based on the number of self-reinforcing patrol paths and the number of balanced patrol paths. The method for calculating influence according to feature 3.

6. The calculation process described above is: Calculated based on the number of pathways including the aforementioned objective factor, The method for calculating influence according to feature 3.

7. The calculation process described above is: Calculated based on the average number of nodes in the aforementioned circuit path, The method for calculating influence according to feature 3.

8. From a directed cyclic graph, extract cyclic paths that include the target objective factor and the focus factor. The aforementioned travel paths are classified into self-reinforcing travel paths where the product of causal coefficients is positive, and balanced travel paths where the product is negative. The strength of the self-reinforcing circulation path is calculated as the strength of the circulation influence of the focus factor on the objective factor in the direction of promotion. An influence calculation program characterized by having a computer perform the processing.

Images