Workflow generation method and workflow generation program

Abstract

To generate workflow that achieves a higher KPI with the existing workflow as reference.SOLUTION: A plurality of candidate workflows obtained by adding change to a portion of the existing workflow are created, a KPI variation vectors are respectively acquired from the existing workflow about the plurality of respective candidate workflows that are created, two or more KPI variation vectors among the plurality of KPI variation vectors are synthesized to generate a plurality of synthetic measures, a weight parameter set such that upper change weight in a selection route becomes higher than lower change weight in the selection route is reflected in the case where a plurality of changes are included in one selection route in a candidate workflow, respective KPI prediction values of the plurality of synthetic measures are calculated, and workflows of the synthetic measures in which the KPI prediction values satisfy a target value are outputted.SELECTED DRAWING: Figure 30

Description

[Technical Field]

[0001] This invention relates to a workflow generation method and a workflow generation program. [Background technology]

[0002] In policy planning, proposals are often based on intuition, experience, and assumptions, making it difficult to achieve key performance indicator (KPI) targets. Therefore, there is a demand to generate and present policy proposals that can achieve targets based on evidence-based KPI predictions at the policy planning stage.

[0003] Conventionally, for example, technologies have been proposed to predict the KPIs of policy candidates from integrated data consisting of various types, thereby supporting decision-making in policy formulation. For example, integrated data is prepared in advance, including the history of daily health information, diagnostic results, lifestyle habits, and test results for each individual resident of the region.

[0004] Furthermore, as a policy KPI prediction, we predict the probability (classified into three categories: high, medium, and low) that each person in the target area will become obese within a specified period if the proposed policy is implemented.

[0005] In setting policy candidates, for example, the policy goal could be set as "reduce the obesity rate among adults in the target area to 25%", the policy candidate could be set as "provide exercise and nutrition guidance to young residents of the area by internal medicine physicians", and the KPI could be set as "the obesity rate among adults in the area". [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2022-14106 [Patent Document 2] Japanese Patent Publication No. 2005-332270 [Patent Document 3] Japanese Patent Application Laid-Open No. 2017-208035 [Patent Document 4] Japanese Patent Application Laid-Open No. 2021-72022 [Patent Document 5] Specification of Japanese Patent No. 6799313 [Non-Patent Document]

[0007] [Non-Patent Document 1] Konstantinos Moutselos; Dimosthenis Kyriazis; Ilias Maglogiannis, “A web based modular environment for assisting health policy making utilizing big data analytics”, [online], July 23 - 25, 2018, IEEE, [searched on June 15, 2022], Internet <URL:https: / / ieeexplore.ieee.org / document / <8633625> [Summary of the Invention] [Problems to be Solved by the Invention]

[0008] However, in such a conventional decision-making support method for policy formulation, after predicting the KPI for a policy, when it is found that the policy cannot achieve the goal, it is difficult to generate a new policy. For example, in the original policy candidate, if the predicted result of the KPI "adult obesity prevalence rate" is 30%, and it does not reach the KPI target of 25%, the policy maker has to manually reconsider the policy candidate again, which is cumbersome.

[0009] Therefore, there is also a method of regenerating new measures when the goal cannot be achieved. For example, a synthetic measure is created by combining multiple measure candidates, and all possible combinations of this synthetic measure are enumerated. For example, when there are N measure candidates, 2 to the power of N synthetic measures are created. Then, KPI prediction is sequentially performed for each synthetic measure, and when there is a measure candidate that can achieve the goal, the said measure candidate is adopted.

[0010] However, in this method, as the number of measure candidates increases, the combinations of synthetic measures increase exponentially. Therefore, there is a problem that the computational load increases in order to perform simulations of KPI predictions for all synthetic measures.

[0011] In particular, when there are a large number of parameters that can be changed in a measure, the combinations of synthetic measures become enormous, and the computational load for executing KPI prediction simulations increases to an unrealistic level.

[0012] In one aspect, the present invention aims to generate a workflow that realizes a higher KPI based on an existing workflow.

Means for Solving the Problem

[0013] Therefore, this workflow generation method creates a plurality of candidate workflows by modifying a part of an existing workflow that defines a plurality of conditional branches and the follow-up content reached at each branch destination. For each of the plurality of created candidate workflows, a KPI change vector from the existing workflow is obtained. Two or more of the plurality of KPI change vectors are combined to generate a plurality of combined measures. When a plurality of the changes are included in one selection route in the candidate workflow, a weight parameter is set such that the weight of the upper change in the selection route is higher than the weight of the lower change in the selection route. The computer executes a process of calculating the KPI prediction value of each of the plurality of combined measures, and outputting the workflow of the combined measure whose KPI prediction value satisfies the target value.

Effect of the Invention

[0014] According to one embodiment, a workflow that realizes a higher KPI can be generated based on an existing workflow.

Brief Description of the Drawings

[0015] [Figure 1] It is a diagram for explaining a KPI prediction method of a combined measure as related art. [Figure 2] It is a diagram showing a CKD (Chronic Kidney Disease) follow-up system in a specific health check using a workflow. [Figure 3] It is a diagram for explaining a method of generating a combined measure using the KPI prediction shown in FIG. 1. [Figure 4] It is a diagram exemplifying a node assignment result and a KPI provisional prediction result for the method of generating a combined measure exemplified in FIG. 3. [Figure 5] It is a diagram exemplifying a node assignment result and a KPI provisional prediction result when the influence ranges of perturbations overlap in the method of generating a combined measure using KPI prediction. [Figure 6]This block diagram shows an example of a computer hardware (HW) configuration that realizes the functionality of a workflow generation device as an example of an embodiment. [Figure 7] This diagram illustrates the functional configuration of a workflow generation device as an example of an embodiment. [Figure 8] This is a diagram illustrating a policy workflow in a workflow generation device as an example of an embodiment. [Figure 9] This figure illustrates a method for adding perturbations in a workflow generation device as an example of an embodiment. [Figure 10] This diagram illustrates a method for generating policy candidates by the policy candidate generation unit of a workflow generation device, as an example of an embodiment. [Figure 11] This figure illustrates multiple policy candidates generated by the policy candidate generation unit of a workflow generation device as an example of an embodiment. [Figure 12] This diagram illustrates the processing of the policy candidate KPI prediction unit of a workflow generation device as an example of an embodiment. [Figure 13] This diagram illustrates the status of each individual and the intervention node assignment results for each candidate measure in a workflow generation device, as an example of an embodiment. [Figure 14] This diagram illustrates the predicted KPI results for each individual for each proposed measure in a workflow generation device, as an example of an embodiment. [Figure 15] This diagram illustrates the method for calculating KPI prediction values ​​by the policy candidate KPI prediction unit of a workflow generation device, as an example of an embodiment. [Figure 16] This figure shows an example of the final KPI prediction values ​​predicted by the policy candidate KPI prediction unit of the workflow generation device, as an example of an embodiment. [Figure 17] This figure illustrates a method for generating fluctuation vectors by the policy candidate KPI prediction unit of a workflow generation device, as an example of an embodiment. [Figure 18]This figure illustrates the fluctuation vector generated by the policy candidate KPI prediction unit of a workflow generation device as an example of an embodiment. [Figure 19] This diagram illustrates the functional configuration of the provisional KPI prediction unit for synthesized measures in a workflow generation device, as an example of an embodiment. [Figure 20] This diagram illustrates the processing of the provisional KPI prediction unit for synthesized measures in a workflow generation device, as an example of an embodiment. [Figure 21] This figure illustrates a combination pattern of weights for perturbations in a workflow generation device as an example of an embodiment. [Figure 22] This figure illustrates the processing performed by the weight adjustment parameter calculation unit in a workflow generation device as an example of an embodiment. [Figure 23] This diagram illustrates the flow depth in a policy workflow in a workflow generation device as an example of an embodiment. [Figure 24] This figure shows an example of a weight pattern and adjustment parameters in a workflow generation device as an example of an embodiment. [Figure 25] This figure shows the synthesis strategy after weight adjustment using weight adjustment parameters in a workflow generation device as an example of an embodiment. [Figure 26] This is a diagram illustrating the processing of the combination determination unit in a workflow generation device as an example of an embodiment. [Figure 27] This figure illustrates KPI targets in a workflow generation device as an example of an embodiment. [Figure 28] This diagram illustrates the processing of the synthetic policy KPI prediction unit in a workflow generation device as an example of an embodiment. [Figure 29] This figure illustrates the output information generated by the output control unit of a workflow generation device as an example of an embodiment. [Figure 30] This is a flowchart illustrating the overview of processing in a workflow generation device as an example of an embodiment. [Figure 31]This is a flowchart illustrating the details of the processing in a workflow generation device as an example of an embodiment. [Modes for carrying out the invention]

[0016] (I) Related technologies The following methods can be considered for predicting the KPIs of the aforementioned composite measures. Figure 1 is a diagram illustrating the KPI prediction method for composite measures as a related technology.

[0017] First, KPI predictions are made for the existing measure (1) and the candidate measures to be combined (up to N). In Figure 1, the symbol A represents an example of the KPI prediction results, where the KPI prediction results for the one existing measure and the two candidate measures (candidate measures 1 and 2) are plotted in a coordinate space with KPI#1 on the horizontal axis and KPI#2 on the vertical axis.

[0018] Next, we calculate a vector (called the KPI fluctuation vector) that shows the difference between the predicted KPI values ​​of existing measures and the predicted KPI values ​​of each candidate measure.

[0019] In Figure 1, the symbol B represents the KPI fluctuation vectors showing the difference between policy candidates 1 and 2 (shown by symbol A) and existing policies.

[0020] Then, the predicted value of the combined measure is obtained by simply adding the KPI fluctuation vectors of each candidate measure to the predicted KPI values ​​of the existing measures.

[0021] In Figure 1, symbol C represents a composite policy generated (predicted) by summing the KPI fluctuation vectors of each policy candidate shown in symbol B.

[0022] Here, a policy may be represented as a workflow combining multiple conditional distributions and multiple nodes. Furthermore, a workflow representing a policy may be called a policy workflow. A policy workflow may include elements other than conditional branches and nodes.

[0023] Figure 2 is a diagram illustrating the CKD (Chronic Kidney Disease) follow-up system in specific health checkups using a workflow.

[0024] In Figure 2, symbol A represents the CKD follow-up system diagram for specific health checkups, and symbol B represents the follow-up system diagram shown in symbol A as a workflow.

[0025] In specific health checkups (specific screenings), for example, "the eGFR test value is a predetermined threshold (e.g., 50 ml / min / 1.73 m³)." 2 Each decision branch based on the test result, such as "is it less than )" or "is urine protein 2+ or higher," corresponds to a conditional branch in the workflow.

[0026] Furthermore, the "treatment by a nephrologist," "insurance guidance by the attending physician," and "treatment by a diabetes specialist" identified as results of each decision branch correspond to nodes in the workflow. Since these nodes involve intervention by either a nephrologist, attending physician, or diabetes specialist, each node can also be called an intervention node.

[0027] For each individual targeted by the initiative, the system determines which node to assign to them by sequentially following multiple conditional branches based on their individual status data within the initiative workflow. Each node corresponds to the follow-up action reached at the branching point in the workflow.

[0028] Figure 3 illustrates the method for generating composite strategies using the KPI forecast shown in Figure 1.

[0029] In Figure 3, symbol A represents an example of an existing policy. Symbol B represents policy candidate 1, symbol C represents policy candidate 2, and symbol D represents a combined policy.

[0030] Policy candidate 1, indicated by symbol B, is an existing policy with perturbation A added, and policy candidate 2, indicated by symbol C, is an existing policy with perturbation B added. The composite policy indicated by symbol D is a combination of policy candidate 1 and policy candidate 2.

[0031] A perturbation is a predetermined change applied to a policy workflow within a defined scope. Perturbations may be applied to elements constituting the policy workflow (e.g., conditional branches, nodes). The predetermined change may be, for example, increasing the number of criteria parameters in a condition.

[0032] As a prerequisite, in order to predict KPIs using a policy workflow, the workflow must be applied to all policy targets, the number of people to be assigned to each node must be calculated, and KPI predictions are made based on these node assignment results. For example, if the number of people assigned to the obesity guidance node is small, the KPI for obesity prevalence will be high.

[0033] Figure 4 illustrates the node allocation results and provisional KPI forecast results for the composite strategy generation method exemplified in Figure 3.

[0034] Based on the above assumptions, the node allocation results for policy candidate 1 and policy candidate 2 are obtained, as shown in Figure 4. In Figure 4, under the label B, the change in the number of people assigned to a node due to perturbation A (assignment result variation) is shown by a number in parentheses. Similarly, under the label C, the change in the number of people assigned to a node due to perturbation B (assignment result variation) is shown by a number in parentheses.

[0035] In the example shown in Figure 4, the node allocation result for the composite policy can be obtained by summing the node allocation result of the existing policy with the difference in node allocation results for each policy candidate 1 and 2.

[0036] Therefore, it is appropriate to determine the predicted KPI values ​​for the combined measures by adding the predicted KPI values ​​for existing measures with the KPI fluctuation vectors of measure candidates 1 and 2.

[0037] However, if the scopes of influence of perturbations overlap, this prediction method will not work correctly because the higher-level perturbation will have a greater influence, while the lower-level perturbation will have less influence.

[0038] Figure 5 illustrates the node assignment results and provisional KPI prediction results when the scope of perturbation influences overlap in a composite strategy generation method using KPI prediction.

[0039] In the example shown in Figure 5, perturbation B of policy candidate 2 is included within the scope of influence of perturbation A of policy candidate 1, resulting in an overlap between the scope of influence of perturbation A of policy candidate 1 and the scope of influence of perturbation B of policy candidate 2. When the scopes of influence overlap, the same target group of the policies may pass through both scopes of influence.

[0040] In a workflow, the upstream side is referred to as the upper level, and the downstream side as the lower level. In the example shown in Figure 5, perturbation A of policy candidate 1 corresponds to the upper level, and perturbation B of policy candidate 2 corresponds to the lower level.

[0041] When combining policy candidate 1 and policy candidate 2, the node assignment result of the combined policy will be dominated by the node assignment result of policy perturbation 1, which is higher in rank. Therefore, it is not simply a sum of the differences in the node assignment results of each policy candidate 1 and 2.

[0042] Therefore, the KPI forecast values ​​for the combined measures cannot be simply added together; adding the KPI forecast values ​​of existing measures with the KPI fluctuation vectors of candidate measures 1 and 2 may lead to incorrect estimates.

[0043] Therefore, in the information processing device 1 as an example of this embodiment, when candidate measures to be synthesized are synthesized, if there is an overlap in the scope of influence between each perturbation, the lower-level perturbation is affected by the higher-level perturbation, thereby enabling more accurate KPI prediction.

[0044] (II) Embodiment The following describes embodiments of the workflow generation method and workflow generation program with reference to the drawings. However, the embodiments shown below are merely illustrative, and there is no intention to exclude various modifications or applications of technologies not explicitly shown in the embodiments. In other words, these embodiments can be implemented with various modifications without departing from their spirit. Furthermore, each figure is not intended to represent only the components shown in the figure, but may include other functions, etc.

[0045] Workflow generation device 1 generates a workflow that can achieve higher KPIs based on the workflow of an existing initiative (existing initiative workflow).

[0046] (A) Hardware configuration example In one embodiment, the workflow generation device 1 may be a virtual server (VM; Virtual Machine) or a physical server. Furthermore, the functions of the workflow generation device 1 may be implemented by one computer or by two or more computers. In addition, at least a portion of the functions of the workflow generation device 1 may be implemented using hardware (HW) resources and network (NW) resources provided by a cloud environment.

[0047] Figure 6 is a block diagram showing an example of the hardware (HW) configuration of a computer 10 that implements the functions of a workflow generation device 1 as an example of an embodiment. Therefore, if multiple computers are used as hardware resources to realize the functions of the workflow generation device 1, each computer may have the hardware configuration illustrated in Figure 6.

[0048] As shown in Figure 6, the computer 10 may, as an example of its hardware configuration, include a processor 10a, a graphics processing unit 10b, a memory 10c, a storage unit 10d, an IF (Interface) unit 10e, an IO (Input / Output) unit 10f, and a read unit 10g.

[0049] Processor 10a is an example of an arithmetic processing unit that performs various control and calculations. Processor 10a may be connected to each block in the computer 10 via bus 10j so as to be able to communicate with each other. Processor 10a may be a multiprocessor containing multiple processors, a multicore processor having multiple processor cores, or a configuration having multiple multicore processors.

[0050] Examples of processor 10a include integrated circuits (ICs) such as CPUs, MPUs, APUs, DSPs, ASICs, and FPGAs. Note that two or more combinations of these integrated circuits may be used as processor 10a. CPU stands for Central Processing Unit, MPU for Micro Processing Unit, APU for Accelerated Processing Unit, DSP for Digital Signal Processor, ASIC for Application Specific IC, and FPGA for Field-Programmable Gate Array.

[0051] The graphics processing unit 10b controls the screen display for output devices such as monitors, which are part of the I / O unit 10f. The graphics processing unit 10b can be various types of processing units, such as integrated circuits (ICs) including GPUs (Graphics Processing Units), APUs, DSPs, ASICs, or FPGAs.

[0052] Memory 10c is an example of hardware that stores various data and program information. Examples of memory 10c include volatile memory such as DRAM (Dynamic Random Access Memory) and non-volatile memory such as PM (Persistent Memory), or both.

[0053] The storage unit 10d is an example of hardware that stores various data and program information. Examples of storage units 10d include magnetic disk devices such as HDDs (Hard Disk Drives), semiconductor drive devices such as SSDs (Solid State Drives), and various storage devices such as non-volatile memory. Examples of non-volatile memory include flash memory, SCM (Storage Class Memory), and ROM (Read Only Memory).

[0054] The storage unit 10d may store a program 10h (workflow generation program) that implements all or part of the various functions of the computer 10.

[0055] For example, the processor 10a of the workflow generation device 1 can realize the workflow generation function described later by loading the program 10h stored in the storage unit 10d into the memory 10c and executing it. In addition, the storage unit 10d may store various data generated during the processing process by each part (see Figure 7) that realizes the functions of the workflow generation device 1.

[0056] The IF unit 10e is an example of a communication interface that controls the connection and communication between this computer 10 and other computers. For example, the IF unit 10e may include an adapter compliant with LAN (Local Area Network) such as Ethernet®, or optical communication such as FC (Fibre Channel). The adapter may support wireless, wired, or both communication methods.

[0057] For example, the workflow generation device 1 may be connected to other information processing devices (not shown) via the IF unit 10e and a network so as to be able to communicate with each other. The program 10h may be downloaded from the network to the computer 10 via the communication IF and stored in the storage unit 10d.

[0058] The I / O unit 10f may include either an input device or an output device, or both. Examples of input devices include keyboards, mice, and touch panels. Examples of output devices include monitors, projectors, and printers. The I / O unit 10f may also include a touch panel that integrates an input device and a display device. The output device may be connected to the graphics processing unit 10b.

[0059] The reading unit 10g is an example of a reader that reads data and program information recorded on the recording medium 10i. The reading unit 10g may include a connection terminal or device to which the recording medium 10i can be connected or inserted. Examples of the reading unit 10g include an adapter compliant with USB (Universal Serial Bus), a drive device for accessing a recording disk, and a card reader for accessing flash memory such as an SD card. The recording medium 10i may store a program 10h, and the reading unit 10g may read the program 10h from the recording medium 10i and store it in the storage unit 10d.

[0060] Examples of recording media 10i include non-temporary computer-readable recording media such as magnetic / optical discs and flash memory. Examples of magnetic / optical discs include flexible discs, CDs (Compact Discs), DVDs (Digital Versatile Discs), Blu-ray discs, and HVDs (Holographic Versatile Discs). Examples of flash memory include semiconductor memory such as USB memory and SD cards.

[0061] The hardware configuration of computer 10 described above is illustrative. Therefore, the addition or deletion of hardware within computer 10 (for example, adding or deleting arbitrary blocks), division, integration in any combination, or addition or deletion of buses may be performed as appropriate.

[0062] (B) Example of functional configuration Figure 7 is a diagram illustrating the functional configuration of the workflow generation device 1 as an example of an embodiment.

[0063] As shown in Figure 7, the workflow generation device 1 may, for example, include the functions of an existing policy acquisition unit 101, a policy candidate generation unit 102, a policy candidate KPI prediction unit 103, a composite policy provisional KPI prediction unit 104, a composite policy actual KPI prediction unit 105, and an output control unit 106. These functions may be implemented by the hardware of the computer 10 (see Figure 6).

[0064] The existing policy acquisition unit 101 acquires existing policies. The existing policy acquisition unit 101 may acquire existing policies by reading them from a predetermined storage area of ​​the storage unit 10d, for example, which have been created using a known method in advance. Alternatively, the existing policy acquisition unit 101 may acquire existing policies from another computer connected via the IF unit 10e.

[0065] A policy is represented in the form of a workflow with multiple conditional branches and multiple nodes (intervention nodes). A workflow representing a policy can be called a policy workflow or policy flow. Existing policies are also represented by policy workflows.

[0066] For each individual targeted by the intervention, the intervention workflow is determined by sequentially following multiple conditional branches based on the individual's status data, thereby deciding which intervention node to assign to that individual.

[0067] Figure 8 is a diagram illustrating the policy workflow in the workflow generation device 1 as an example of an embodiment.

[0068] In Figure 8, symbol A represents an example of an existing policy workflow. This example of an existing policy workflow includes four conditional branches L1 to L4 and four nodes (intervention nodes) #1 to #4.

[0069] The policy workflow has a tree structure, and the founding node (upper side in the diagram), which has no parent, can be called the upstream or upper side, while the terminal node (lower side in the diagram), which has no children, can be called the downstream or lower side.

[0070] In Figure 8, symbol B indicates the flow definition table corresponding to the existing policy workflow indicated by symbol A. The flow definition table is information that represents the structure of the policy workflow, and it specifies all information regarding conditional branches included in the policy workflow in a table format.

[0071] The flow definition table contains descriptions of branch destinations and descriptions of branch conditions. In the flow definition table shown by symbol B in Figure 8, the description of branch destinations includes, for example, that if conditional branch L1 is True, the process proceeds to conditional branch L2, and if conditional branch L1 is False, the process proceeds to conditional branch L3.

[0072] In the flow definition table indicated by symbol B in Figure 8, the description of the branching conditions states, for example, that in conditional branch L1, it is determined to be True if the value of eGFR is less than 50 (eGFR<50).

[0073] In this workflow generation device 1, information on policy workflows may be managed using such a flow definition table.

[0074] The existing policy acquisition unit 101 may acquire the existing policy workflow along with the flow definition table of the said existing policy workflow. Alternatively, the existing policy acquisition unit 101 may create a flow definition table based on the existing policy workflow.

[0075] The policy candidate generation unit 102 generates policy candidates by adding perturbations to the existing policy workflow.

[0076] Figure 9 illustrates a method for adding perturbations in the workflow generation device 1 as an example of an embodiment.

[0077] Figure 9 shows four types of perturbation methods, as shown in (a) to (d) below.

[0078] (a) Change the condition of the conditional branch (for example, change eGFR<50 to eGFR<40) (b) Reduce or increase the number of conditional branches (c) Change the type of intervention at the intervention node (for example, change health guidance to home visit guidance) (d) Reduce or increase the number of intervention nodes

[0079] The policy candidate generation unit 102 generates multiple policy candidates by applying such perturbations to conditional branches and nodes in the existing policy workflow. The method of applying perturbations to each conditional branch or node is defined in advance.

[0080] Figure 10 is a diagram illustrating the method for generating policy candidates by the policy candidate generation unit 102 of the workflow generation device 1, as an example of an embodiment.

[0081] The policy candidate generation unit 102 creates policy candidates by adding, for example, the perturbation of pattern (a) above to an existing policy workflow. In this process, variations in the perturbation are introduced by changing the conditional branch to which the perturbation is added, or by changing the threshold for the same conditional branch, thereby generating multiple types of policy candidates.

[0082] Furthermore, when the policy candidate generation unit 102 creates policy candidates by adding the perturbation of pattern (b) described above, it may create variations in the perturbation by changing the conditional branch to be reduced, thereby generating multiple types of policy candidates. Similarly, the policy candidate generation unit 102 may create policy candidates by adding the perturbations of pattern (c) and pattern (d) described above.

[0083] Figure 11 is a diagram illustrating multiple policy candidates generated by the policy candidate generation unit 102 of the workflow generation device 1 as an example of an embodiment.

[0084] Figure 11 shows the flow definition tables for each of the multiple policy candidates. Each of the policy candidates is assigned a unique policy number, and Figure 11 shows four policy candidates with policy numbers 1 through 4. In Figure 11, policy candidate with policy number 0 is an existing policy workflow.

[0085] For example, in the case of policy candidate No. 1, the threshold of branching condition L1 has been changed from 50 to 40 compared to the existing policy candidate (policy No. 0). Also, in the case of policy candidate No. 4, branching condition L9 has been added compared to the existing policy candidate (policy No. 0).

[0086] The policy candidate generation unit 102 creates multiple candidate workflows (policy candidates) by perturbing (modifying) a part of an existing workflow that defines multiple conditional branches and intervention nodes (follow-up content) reached at each branch destination.

[0087] The policy candidate information generated by the policy candidate generation unit 102 may be stored in a predetermined storage area of ​​the storage unit 10d.

[0088] The policy candidate KPI prediction unit 103 predicts KPIs for each policy candidate generated by the policy candidate generation unit 102.

[0089] Figure 12 is a diagram illustrating the processing of the policy candidate KPI prediction unit 103 of the workflow generation device 1 as an example of an embodiment.

[0090] The policy candidate KPI prediction unit 103 first determines which intervention node to assign to each individual by following conditional branches for each policy candidate based on each individual's status data and identifying the intervention node (follow-up content) to reach.

[0091] In Figure 12, symbol A represents the process of determining which intervention node to assign an individual to within the proposed policy.

[0092] The policy candidate KPI prediction unit 103 identifies the node (intervention node) to be assigned to a specific individual by sequentially making selections in accordance with conditional branching from the upstream to the downstream side of the policy candidate, as exemplified by symbol A in Figure 12.

[0093] Next, the policy candidate KPI prediction unit 103 performs various KPI predictions based on the individual's status data and the intervention node assigned to that individual.

[0094] In Figure 12, the symbol B represents the process of making KPI predictions at the individual level.

[0095] For example, if the KPI is the rate of new CKD cases, the policy candidate KPI prediction unit 103 predicts the status (s1,...,s) for each individual. N ) and intervention nodes (T1,...,T M The data is input into a predetermined prediction model (KPI prediction model) and the new CKD incidence rate is output. A prediction model may be set up for each KPI.

[0096] Furthermore, the construction of a predictive model may be carried out by applying machine learning algorithms such as Bayesian model training or deep learning to historical machine learning data.

[0097] The prediction model may be, for example, a deep learning model (deep neural network). The neural network may be a hardware circuit, or it may be a virtual network of software connecting layers virtually constructed on a computer program by a processor 10a (see Figure 6), etc. Figure 13 is a diagram illustrating the status of each individual and the intervention node assignment results for each policy candidate in the workflow generation device 1, as an example of an embodiment. Figure 14 is a diagram illustrating the predicted KPI results for each individual for each policy candidate in the workflow generation device 1, as an example of an embodiment.

[0098] Figures 13 and 14 show an example where there are 4 policy candidates and 6,875 individuals.

[0099] Figure 13 shows the values ​​corresponding to each individual's branching condition and the assignment results of intervention nodes for each measure.

[0100] The individuals in each measure shown in Figure 14 correspond to the individuals in each measure in Figure 13. When the node assignment results change in a measure, the KPI values ​​may also change. For example, in Figure 13, the individual with personal ID 6875 is assigned to intervention node #4 in the existing measure (measure No. 0), but is assigned to intervention node #5 in measure No. 4.

[0101] As a result, as shown in Figure 14, for the individual with personal ID 6875, under the existing measure (Measure No. 0), the KPI1_CKD incidence rate is 0.87 and the KPI2_intervention node cost is 9700, whereas under Measure No. 4, the KPI1_CKD incidence rate becomes 0.79 and the KPI2_intervention node cost becomes 10500.

[0102] Figure 15 is a diagram illustrating the method for calculating KPI prediction values ​​by the policy candidate KPI prediction unit 103 of the workflow generation device 1, as an example of an embodiment.

[0103] The policy candidate KPI prediction unit 103 obtains predicted KPI values ​​by inputting information on the individual's condition and intervention node for all individuals for each policy into the KPI prediction model. A separate KPI prediction model is prepared for each KPI.

[0104] In the example shown in Figure 15, a KPI#1 prediction model that generates a predicted value for KPI#1 and a KPI2 prediction model that generates a predicted value for KPI#2 are shown.

[0105] The policy candidate KPI prediction unit 103 calculates the KPI#1 predicted value for the policy candidate by calculating the average of the KPI#1 predicted values ​​for all individuals calculated for the same policy candidate. Similarly, the policy candidate KPI prediction unit 103 calculates the KPI#2 predicted value for the policy candidate by calculating the average of the KPI#2 predicted values ​​for all individuals calculated for the same policy candidate.

[0106] Figure 16 shows an example of the final KPI prediction values ​​predicted by the policy candidate KPI prediction unit 103 of the workflow generation device 1 as an example of an embodiment.

[0107] As shown in Figure 16, the policy candidate KPI prediction unit 103 calculates the average value of the predicted KPI for each type of KPI for each policy candidate. The policy candidate KPI prediction unit 103 calculates the predicted KPI value (average value) for all types of KPIs for all policy candidates. Furthermore, the policy candidate KPI prediction unit 103 calculates the difference between the predicted KPI value of an existing policy and the predicted KPI value of each policy candidate for all policy candidates, and obtains the fluctuation vector for each perturbation. The vector representing the difference between the predicted KPI value of an existing policy and the predicted KPI value of each policy candidate can be called the difference vector.

[0108] Figure 17 is a diagram illustrating the method for generating fluctuation vectors by the policy candidate KPI prediction unit 103 of the workflow generation device 1 as an example of an embodiment.

[0109] In Figure 17, two examples of KPIs are shown, with the existing measures and candidate measures 1-3 arranged in a two-dimensional coordinate space where the horizontal axis represents the predicted value of KPI #1 and the vertical axis represents the predicted value of KPI #2. Four difference vectors are shown, generated by taking the difference between the predicted KPI value of the existing measures and the predicted KPI values ​​of each of the candidate measures 1-4.

[0110] Figure 18 is a diagram illustrating the fluctuation vector generated by the policy candidate KPI prediction unit 103 of the workflow generation device 1 as an example of an embodiment.

[0111] In Figure 18, for each of the policy candidates 1 to 4, the change vector of KPI #1 compared to the existing policy (change vector_KPI#1) and the change vector of KPI #2 compared to the existing policy (change vector_KPI#2) are shown.

[0112] The policy candidate KPI prediction unit 103 calculates the KPI fluctuation vector from the existing workflow (the difference between the predicted KPI value of the existing workflow and the predicted KPI value of each candidate workflow) for each of the multiple candidate workflows (policy candidates).

[0113] The policy candidate KPI prediction unit 103 stores the information of the fluctuation vector of each predicted KPI value for each policy candidate that it has generated in a predetermined storage area of ​​the storage unit 10d.

[0114] Figure 19 illustrates the functional configuration of the Synthetic Policy Provisional KPI Prediction Unit 104 of the Workflow Generation Device 1 as an example of an embodiment.

[0115] The provisional KPI prediction unit 104 generates a composite policy based on the KPI prediction values ​​of each policy candidate calculated by the policy candidate KPI prediction unit 103. The provisional KPI prediction unit 104 generates multiple composite policies by combining two or more KPI fluctuation vectors from among multiple KPI fluctuation vectors.

[0116] The combined policy provisional KPI prediction unit 104, when combining policy candidates to be combined, predicts the KPIs by giving higher weight to the policy candidate whose perturbation occurs upstream of the policy when there is an overlap in the scope of influence between multiple perturbations.

[0117] When a single selection route in a policy workflow includes multiple perturbations, there is an overlap in the scope of influence between these perturbations.

[0118] As shown in Figure 19, the composite policy provisional KPI prediction unit 104 has the functions of a weight combination enumeration unit 201, a weight adjustment parameter calculation unit 202, a provisional KPI prediction calculation unit 203, and a combination determination unit 204.

[0119] Figure 20 is a diagram illustrating the processing of the Synthetic Policy Provisional KPI Prediction Unit 104 of the Workflow Generation Device 1 as an example of an embodiment.

[0120] In Figure 20, symbol A indicates the process by which the composite policy provisional KPI prediction unit 104 generates a composite policy by weighting and summing each perturbation for existing policies. Symbol B indicates a composite policy that satisfies the KPI target.

[0121] A target threshold #1 is set for KPI #1, and a target threshold #2 is set for KPI #2. The area where both target thresholds #1 and #2 are satisfied can be called the KPI target achievement area.

[0122] The composite measure provisional KPI prediction unit 104 generates composite measures that satisfy both target thresholds #1 and #2, i.e., measures that fall within the KPI target achievement area, by setting weights (0 or 1 in this embodiment) for the candidate measures.

[0123] In this workflow generation device 1, a composite strategy is generated that satisfies all KPI targets by weighting and summing each perturbation applied to existing strategies. Therefore, the composite strategy provisional KPI prediction unit 104 calculates provisional values ​​for the KPI prediction of the composite strategy.

[0124] In this workflow generation device 1, the provisional predicted KPI Y of the composite measure is represented by a weighted sum of the existing measure and the fluctuating vector, as shown in equation (1) below.

[0125] Y = X0 + W1 × ΔX1 + ... + W N × ΔX N ...(1) Here, X0 is the predicted KPI value of the existing measure. W1 is the weight of perturbation 1. ΔX1 is the fluctuation vector of perturbation 1. W N ΔX is the weight of the perturbation N. N This is the variation vector of the perturbation N.

[0126] The weight combination enumeration unit 201 uses the weights W1 to W used in the provisional predicted KPI Y of the composite measure represented by equation (1) above. N This generates combination patterns. In other words, the weight combination enumeration unit 201 generates combination patterns of weights to apply to the perturbation. These combination patterns of weights can also be called weight patterns.

[0127] Figure 21 illustrates an example of a combination pattern of weights for perturbations in the workflow generation device 1 as an example of an embodiment.

[0128] In the example shown in Figure 21, multiple weight patterns (for example, weight patterns 1 to 4) are associated with perturbations 1 to 4. A weight pattern is a combination of weight values ​​set for each of perturbations 1 to 4. In the example shown in Figure 21, the weight of each perturbation is represented by either 0 or 1, with a weight of 1 indicating that the perturbation is used in the composite policy, and a weight of 0 indicating that the perturbation is not used in the composite policy.

[0129] For example, in weight pattern 1, perturbations 1 and 4 are used in the composite policy, while perturbations 2 and 3 are not. Also, for example, in weight pattern 4, only perturbation 4 is used in the composite policy, while perturbations 1-3 are not used.

[0130] The various weight combination patterns illustrated in Figure 21 represent candidate perturbations that can be incorporated into existing policy workflows. These weight combination patterns can also be referred to as weight combination candidates.

[0131] As described above using FIG. 4, when generating a composite measure and applying a plurality of perturbations to an existing measure workflow, if the influence ranges of the respective perturbations do not overlap, the predicted KPI value of the composite measure can be obtained by adding the predicted KPI value of the existing measure and the KPI change vector of the measure candidate.

[0132] In contrast, as described above using FIG. 5, when generating a composite measure and applying a plurality of perturbations to an existing measure workflow, if the influence ranges of the respective perturbations overlap, the predicted KPI value of the composite measure cannot be obtained by simply adding the predicted KPI value of the existing measure and the KPI change vector of the measure candidate.

[0133] Therefore, when generating a composite measure and applying a plurality of perturbations to an existing measure workflow, if the influence ranges overlap among the plurality of perturbations, the weight adjustment parameter calculation unit 202 sets a parameter (weight adjustment parameter) for the above-described weights based on the relationship between the perturbations on the existing measure workflow.

[0134] The weight adjustment parameter calculation unit 202 sets the weight adjustment parameter so that the weight of the upper perturbation on the measure workflow has a greater influence than the weight of the lower perturbation.

[0135] FIG. 22 is a diagram for explaining the processing by the weight adjustment parameter calculation unit 202 in the workflow generation device 1 as an example of the embodiment.

[0136] The weight adjustment parameter calculation unit 202 determines the weight adjustment parameter η i (i = 1,..., N) for a plurality of perturbations according to the following procedures (1) to (6).

[0137] Procedure (1): The weight adjustment parameter calculation unit 202 extracts a perturbation group P0 having a value of 1 or more from among the weight combination candidates.

[0138] In Figure 22, the symbol A indicates a candidate weight combination extracted from among multiple weight combination patterns generated by the weight combination enumeration unit 201. In Figure 22, perturbation 1 and perturbation 2 are weight combinations. polish This corresponds to a perturbation of 1 or greater.

[0139] Procedure (2): The weight adjustment parameter calculation unit 202 identifies whether there is overlap in the scope of influence between each perturbation from the structure of the policy workflow and obtains an overlap matrix.

[0140] In Figure 22, symbol B represents an example of a policy workflow, and in this policy workflow shown by symbol B, all perturbations 1 to 4 are reflected. Also in Figure 22, symbol C represents an overlap matrix showing whether or not there is overlap in the scope of influence between perturbations in the policy workflow shown by symbol B. In the overlap matrix shown by symbol C in Figure 22, for example, perturbation 1 overlaps with all of perturbations 2 to 4, that is, perturbation 1 affects all of perturbations 2 to 4. Also, for example, perturbation 2 affects only perturbation 3.

[0141] The weight adjustment parameter calculation unit 202 may generate a duplicate matrix based on the policy workflow using known methods, and a detailed explanation of this method is omitted.

[0142] Furthermore, the weight adjustment parameter calculation unit 202 refers to this overlap matrix to extract a submatrix P1 corresponding to the perturbation group P0. In Figure 22, the symbol D indicates the submatrix P1 corresponding to the perturbation group P0 extracted from the overlap matrix shown in symbol C.

[0143] Procedure (3): The weight adjustment parameter calculation unit 202 refers to the submatrix P1 and sequentially determines whether there is at least one overlap for each perturbation i included in the perturbation group P0. In the example shown by symbol D in Figure 22, the influence range of perturbation 2 is included in (overlaps with) the influence range of perturbation 1.

[0144] Procedure (4): For perturbations i in which there are no overlaps in the influence range, the weight adjustment parameter calculation unit 202 calculates η iLet = 1.

[0145] Procedure (5): On the other hand, for perturbations i that have at least one overlap, the weight adjustment parameter calculation unit 202 sets a higher parameter value for perturbations higher up in the policy workflow and a lower parameter value for perturbations lower down.

[0146] Figure 23 is a diagram illustrating the flow depth in the policy workflow in the workflow generation device 1 as an example of an embodiment.

[0147] In Figure 23, symbol A represents the structure and flow of the policy workflow. deep This indicates the degree. A policy workflow with a tree structure has a hierarchical structure. In such a policy workflow, the higher the hierarchy level, the shallower the flow depth, and the lower the hierarchy level, the deeper the flow depth. In Figure 23, 1 is set as the shallowest flow depth, and 4 is set as the deepest flow depth.

[0148] Procedure (6): The weight adjustment parameter calculation unit 202 sets the parameter η by applying a decay function based on flow depth.

[0149] In Figure 23, the symbol B indicates the decay function based on flow depth. The weight adjustment parameter calculation unit 202 sets higher weight adjustment parameter values ​​for shallower (higher) flow depths in the policy workflow, and lower weight adjustment parameter values ​​for deeper (lower) flow depths.

[0150] When x(i) is the flow depth for the perturbation i in question, the damping function for determining the weight adjustment parameter η can be expressed, for example, by equation (2) shown below. η(i) = exp{-ax(i)} ···(2) In equation (2) above, a is a constant.

[0151] Furthermore, to obtain the final value η′(i), the weight adjustment parameter calculation unit 202 performs normalization as shown in equation (3) below, so that the average value of all η′(i) becomes 1.

number

[0152] Figure 24 shows an example of a weight pattern and adjustment parameters in a workflow generation device 1 as an example of an embodiment.

[0153] In Figure 24, code A represents the combination pattern of weights for the perturbation, which is the same as the combination pattern of weights for the perturbation shown in Figure 21. Code B indicates the result of determining whether there is overlap between perturbations. Code C indicates the calculation result of the weight adjustment parameter.

[0154] Figure 25 shows a composite strategy after weight adjustment using weight adjustment parameters in the workflow generation device 1, as an example of an embodiment.

[0155] In Figure 25, two examples of KPIs are shown, and a composite policy (see symbols a and b) is shown in a two-dimensional coordinate space where the horizontal axis represents the predicted value of KPI#1 and the vertical axis represents the predicted value of KPI#2, by combining a policy candidate with a higher-level perturbation and a policy candidate with a lower-level perturbation with an existing policy.

[0156] Furthermore, in Figure 25, the symbol a represents the composite value when no weight adjustment is performed using the weight adjustment parameter η. In contrast, the symbol b represents the composite value obtained by applying weight adjustment using the weight adjustment parameter η, thereby increasing the weight of the policy candidate vectors for the higher-level perturbation and adding them together.

[0157] The KPI provisional forecast calculation unit 203 calculates provisional forecast values ​​for the KPIs of the composite measures for each KPI.

[0158] The KPI provisional forecast calculation unit 203 calculates the KPI provisional forecast value Y for each KPI based on the following formula (4). (k) Calculate. Y (k) = X0 (k) + η1W1 × ΔX1 (k) + ··· +η N W N × ΔX N (k) ...(4) k is a value that identifies any KPI from among L types of KPIs, and k is a natural number greater than or equal to 1. η1 is the adjustment parameter for perturbation 1, and η N This is the adjustment parameter for the perturbation N.

[0159] The composite strategy provisional KPI prediction unit 104 calculates the KPI prediction values ​​(provisional KPI prediction values) for each of the multiple composite strategies by reflecting a weight parameter η that is set such that the weight (W) of the higher-level perturbation in the selected route is higher than the weight (W) of the lower-level perturbation in the selected route, when a single selected route in the candidate workflow includes multiple perturbations (changes).

[0160] The composite strategy provisional KPI prediction unit 104 calculates the provisional KPI prediction values ​​(KPI prediction values) for each of the multiple composite strategies by weighted summation of the KPI prediction values ​​of the existing workflow (existing strategy) and the KPI fluctuation vector (the difference between the KPI prediction values ​​of the existing workflow and the KPI prediction values ​​of each candidate workflow).

[0161] The combination determination unit 204 determines a weight combination (W1, ..., W) such that all KPI predicted values ​​Y, after weight adjustment by the weight adjustment parameter η calculated by the KPI provisional prediction calculation unit 203, satisfy the respective KPI targets. N This is calculated (determined) by applying combinatorial optimization algorithms, etc.

[0162] Figure 26 is a diagram illustrating the processing of the combination determination unit 204 in the workflow generation device 1 as an example of an embodiment.

[0163] The combination determination unit 204 checks whether the provisional predicted value Y for all types of KPIs in the combined measures is less than the target value Z for that KPI, that is, whether all KPIs meet their KPI targets.

[0164] Figure 27 is a diagram illustrating KPI targets in a workflow generation device 1 as an example of an embodiment.

[0165] In Figure 27, for example, KPI #1 is the incidence rate of chronic kidney disease (CKD), and its KPI target Z is... (1) This shows an example where the value is 0.30.

[0166] In the example shown in Figure 26, the combination determination unit 204 determines, for example, the provisional predicted value Y of KPI#1 of the combined measures. (1) However, the target value Z for KPI #1 (1) Is it less than (Y (1) < Z (1) ) Check this.

[0167] For example, the combination determination unit 204 determines the provisional predicted value Y of the KPI#L of the combined measures. (L) However, the target value Z of KPI#L (L) Is it less than (Y (L) < Z (L) ) Check this.

[0168] The combination determination unit 204 then determines a weight combination pattern in which all KPIs satisfy the KPI target as the optimal weight combination pattern for generating a composite measure that can achieve the KPI target.

[0169] The combination determination unit 204 may select multiple weight combination patterns as the optimal weight combination pattern for generating a composite measure that can achieve the KPI target.

[0170] The composite policy actual KPI prediction unit 105 creates a composite policy by adding multiple perturbations to an existing policy based on the weight combination pattern determined by the combination determination unit 204, and then performs an actual KPI prediction for the composite policy.

[0171] Then, the combined policy actual KPI prediction unit 105 checks whether all KPI prediction results meet the target conditions (KPI targets) in the actual KPI prediction of the combined policy.

[0172] Figure 28 is a diagram illustrating the processing of the synthesized policy KPI prediction unit 105 in the workflow generation device 1 as an example of an embodiment.

[0173] The combined policy KPI prediction unit 105 checks whether all KPI prediction results for the combined policy meet the target conditions (KPI targets). If there are any KPI prediction results that do not meet the target conditions, it changes the way the perturbation is applied again and causes the policy candidate generation unit 102 to recreate the combined policy.

[0174] In this case, the method of applying perturbation was to predict the composite policy actual KPIs that were close to the target conditions. Strategy You may introduce a perturbation at the starting point.

[0175] For example, the composite strategy KPI prediction unit 105 calculates the difference between the predicted KPI value and the target value for all composite strategies. If there are multiple KPIs, this difference is calculated for each KPI, and the average value is calculated.

[0176] Then, the composite policy whose average difference was smaller than a predetermined threshold, or the composite policy with the smallest difference, to special It is acceptable to set it.

[0177] The composite policy implementation KPI forecasting unit 105 is a composite policy In the plan In contrast hand, By applying multiple patterns of new types of perturbations (see Figures 9 and 10), new policy candidates are generated.

[0178] Furthermore, the combined measure KPI prediction unit 105 checks whether all KPI prediction results for the combined measure meet the target conditions (KPI targets). If it confirms that all KPI prediction results meet the target conditions (KPI targets), it determines the combined measure as an improvement measure and terminates the process.

[0179] The output control unit 106 outputs information on improvement measures determined by the composite measure provisional KPI forecast unit 104. The output control unit 106 may create output information including information on improvement measures and present it to the user (measure planner). The output information is information that visualizes the information on improvement measures. The output control unit 106 may include information on improvement measures in the output information.

[0180] For example, the output control unit 106 may output the output information to a monitor or the like via the graphics processing unit 10b.

[0181] Figure 29 illustrates the output information generated by the output control unit 106 of the workflow generation device 1 as an example of an embodiment.

[0182] The output information illustrated in Figure 29 is, for example, displayed on the monitor (not shown) of an information processing device used by the user, and includes information on existing measures (see code P01) and information on improvement measures (code P11) arranged side by side.

[0183] For information regarding these existing measures and improvement measures, the measure workflow (see codes P02 and P12) is shown, as well as KPI information (see codes P03 and P13) that correlates the target and predicted values ​​of the KPIs.

[0184] This output information allows users to compare and review existing measures with improvement measures.

[0185] In the implementation workflow for improvement measures, visibility may be improved by changing markers or display colors for parts (perturbations) that have been modified from the implementation workflow of existing measures (see symbols P15 and P16).

[0186] Furthermore, in the KPI information for existing measures, visibility may be improved by changing the font or display color of values ​​where the KPI falls below the target value (see symbol P04). Also, in the KPI information for improvement measures, visibility may be improved by changing the marker or display color of predicted values ​​for KPIs that fell below the target value in such existing measures (see symbol P14).

[0187] The output control unit 106 outputs a workflow for a composite measure in which the provisional KPI forecast value (KPI forecast value) satisfies the target value.

[0188] (C) Operation The flowchart (step S) in Figure 30 shows an overview of the processing in the workflow generation device 1, which is an example of an embodiment configured as described above. 0 1~S 0 5) will be explained as follows.

[0189] Step S 0 In step 1, the existing policy acquisition unit 101 acquires existing policies.

[0190] Step S 0 In step 2, the policy candidate generation unit 102 generates multiple policy candidates by adding perturbations to the existing policy workflow.

[0191] Step S 0 In step 3, the policy candidate KPI prediction unit 103 predicts the KPIs for each policy candidate generated by the policy candidate generation unit 102. The policy candidate KPI prediction unit 103 also calculates the difference between the predicted KPI values ​​of existing policies and the predicted KPI values ​​of each policy candidate for all policy candidates, and obtains the fluctuation vectors for each perturbation.

[0192] Step S 0In step 4, the provisional composite policy KPI prediction unit 104 generates a composite policy by synthesizing the fluctuation vectors of each policy candidate based on the KPI prediction values ​​of each policy candidate calculated by the policy candidate KPI prediction unit 103.

[0193] Furthermore, the combined policy provisional KPI prediction unit 104 sets weight adjustment parameters based on the relationships between perturbations in the existing policy workflow, and calculates provisional KPI prediction values ​​for each combined policy by reflecting these weight adjustment parameters. When policy candidates to be combined are combined, if overlaps occur between each perturbation, the combined policy provisional KPI prediction unit 104 performs KPI prediction by increasing the weight of the policy candidate in which the perturbation occurs upstream of the policy when combining the KPI fluctuation vectors. In step S05, the composite policy actual KPI prediction unit 105 creates a composite policy by adding multiple perturbations to an existing policy based on the weight combination pattern determined by the combination determination unit 204, and then performs an actual KPI prediction for the composite policy.

[0194] A composite measure that satisfies all KPI prediction results (KPI targets) will be adopted as the improvement measure.

[0195] Next, the details of the processing in the workflow generation device 1 as an example of an embodiment will be explained according to the flowchart (steps S1-S7, S41-S44) shown in Figure 31. Processes similar to those described above are indicated by the same reference numerals in the figure.

[0196] In step S1, the existing policy acquisition unit 101 acquires existing policies.

[0197] In step S2, the policy candidate generation unit 102 generates multiple policy candidates by adding perturbations to the existing policy workflow.

[0198] In step S3, the policy candidate KPI prediction unit 103 predicts the KPIs for each policy candidate generated by the policy candidate generation unit 102. The policy candidate KPI prediction unit 103 also calculates the difference between the predicted KPI values ​​of existing policies and the predicted KPI values ​​of each policy candidate for all policy candidates, and obtains the fluctuation vectors for each perturbation.

[0199] The provisional KPI forecasting process for the composite measures shown in step S4 includes the processes in steps S41 to S44.

[0200] In step S41, the weight combination enumeration unit 201 generates multiple weight combination patterns to be applied to the perturbation.

[0201] In step S42, the weight adjustment parameter calculation unit 202, when generating a composite policy and applying multiple perturbations to an existing policy workflow, sets parameters for the weights mentioned above based on the relationships between the perturbations in the existing policy workflow if the scope of influence of each perturbation overlaps.

[0202] In step S43, the KPI provisional forecast calculation unit 203 calculates provisional forecast values ​​for the KPIs of the composite measures for each KPI.

[0203] In step S44, the combination determination unit 204 calculates (determines) a weight combination such that all KPI predicted values, after weight adjustment using the weight adjustment parameters calculated by the KPI provisional prediction calculation unit 203, satisfy their respective KPI targets.

[0204] Subsequently, in step S5, the composite policy actual KPI prediction unit 105 creates a composite policy by adding multiple perturbations to an existing policy based on the weight combination pattern determined by the combination determination unit 204, and then performs an actual KPI prediction for the composite policy.

[0205] In step S6, the composite policy actual KPI prediction unit 105 checks whether all KPI prediction results meet the target conditions (KPI targets) in the actual KPI prediction of the composite policy.

[0206] If, as a result of this check, there are any KPI prediction results that do not meet the target conditions (see NO route in step S6), the process returns to step S2, the method of applying perturbations is changed again, and the policy candidate generation unit 102 is instructed to regenerate the composite policy.

[0207] On the other hand, if the verification in step S6 shows that all KPI forecast results meet the target conditions (KPI targets) (see the YES route in step S6), proceed to step S7.

[0208] In step S7, the output control unit 106 outputs information on the improvement measures determined by the combined measure provisional KPI prediction unit 104. After that, the process ends.

[0209] (D) Effects Thus, according to the workflow generation device 1 as an example of the embodiment, when the composite policy provisional KPI prediction unit 104 synthesizes policy candidates to be synthesized, if overlap occurs between each perturbation, it performs KPI prediction by increasing the weight of the policy candidate in which the perturbation occurs upstream of the policy when synthesizing the KPI fluctuation vectors.

[0210] This means that lower-level perturbations in the policy workflow are affected by higher-level perturbations. Ruko By taking these factors into consideration, more accurate KPI predictions become possible.

[0211] Furthermore, the weight adjustment parameter calculation unit 202 uses a decay function to set a higher value for the weight adjustment parameter η for shallower flow depths (upper levels) in the policy workflow, and a lower value for the weight adjustment parameter η for deeper flow depths (lower levels). This ensures that when combining KPI fluctuation vectors, the weight of policy candidates where perturbations occur upstream of the policy becomes higher.

[0212] (E) Others Furthermore, the disclosed technology is not limited to the embodiments described above, and can be implemented in various modified forms without departing from the spirit of this embodiment. Each configuration and process of this embodiment can be selected or combined as needed.

[0213] For example, in the embodiment described above, as shown in Figure 17, there are two types of KPIs, and the difference vector is shown in a two-dimensional coordinate space where the horizontal axis is the predicted value of KPI#1 and the vertical axis is the predicted value of KPI#2. However, it is not limited to this. There may be three types of KPIs, and the coordinate space in which the difference vector is expanded may be a space of three dimensions or more.

[0214] Furthermore, while the above-described embodiment shows an example where the weight W used in the provisional predicted KPI Y of the composite measure is either 0 or 1, it is not limited to this and may be a value other than 0 or 1.

[0215] Furthermore, in the embodiments described above, the method for adding perturbations to the policy workflow is not limited to the method illustrated in Figure 9, and perturbations may be added by other methods.

[0216] Furthermore, the above disclosure makes it possible for those skilled in the art to implement and manufacture this embodiment.

[0217] (F) Note The following additional information is disclosed regarding the embodiments described above. (Note 1) We created several candidate workflows by modifying parts of an existing workflow that defines multiple conditional branches and the follow-up actions to be taken at each branch. For each of the multiple candidate workflows created, calculate the KPI (Key Performance Indicator) change vector from the existing workflow. Multiple combined measures are generated by combining two or more of the aforementioned KPI fluctuation vectors. If a single selected route in the candidate workflow includes multiple changes, the weight parameters are set such that the weight of the higher-level changes in the selected route is higher than the weight of the lower-level changes in the selected route, and the predicted KPI values ​​for each of the multiple composite measures are calculated accordingly. Output the workflow of the composite measure that satisfies the KPI forecast value. A workflow generation method characterized by having a computer perform the processing.

[0218] (Note 2) The process for calculating the predicted KPI values ​​for each of the aforementioned multiple composite measures is to calculate the predicted KPI values ​​by weighted summing the predicted KPI values ​​of the existing workflow and the KPI fluctuation vector. A workflow generation method according to Appendix 1, characterized by including processing.

[0219] (Note 3) The weight parameter is set according to the flow depth of the candidate workflow. A workflow generation method as described in Appendix 1 or 2, characterized by the above.

[0220] (Note 4) We created several candidate workflows by modifying parts of an existing workflow that defines multiple conditional branches and the follow-up actions to be taken at each branch. For each of the multiple candidate workflows created, calculate the KPI (Key Performance Indicator) change vector from the existing workflow. Multiple combined measures are generated by combining two or more of the aforementioned KPI fluctuation vectors. If a single selected route in the candidate workflow includes multiple changes, the weight parameters are set such that the weight of the higher-level changes in the selected route is higher than the weight of the lower-level changes in the selected route, and the predicted KPI values ​​for each of the multiple composite measures are calculated accordingly. Output the workflow of the composite measure that satisfies the KPI forecast value. A workflow generation program characterized by causing the computer to perform the processing.

[0221] (Note 5) The process for calculating the predicted KPI values ​​for each of the aforementioned multiple composite measures is to calculate the predicted KPI values ​​by weighted summing the predicted KPI values ​​of the existing workflow and the KPI fluctuation vector. A workflow generation program as described in Appendix 4, characterized by including processing.

[0222] (Note 6) The weight parameter is set according to the flow depth of the candidate workflow. A workflow generation program as described in Appendix 4 or 5, characterized by the above. [Explanation of Symbols]

[0223] 1. Workflow generation device 10 Computers 10a processor 10b Graphics Processing Unit 10c memory 10d storage section 10e IF section 10f IO section 10g reading unit 10-hour program 10i recording media 10j bus 101 Existing Policy Acquisition Department 102 Measure candidate generation section 103 Policy Candidate KPI Forecasting Department 104 Synthetic Strategy Provisional KPI Forecasting Department 105 Synthetic Strategy Actual KPI Forecasting Department 106 Output control unit 201 Weight combination enumeration section 202 Weight Adjustment Parameter Calculation Unit 203 KPI Provisional Forecast Calculation Unit 204 Combination Determination Section

Claims

1. We created several candidate workflows by modifying parts of an existing workflow that defines multiple conditional branches and the follow-up actions to be taken at each branch. For each of the multiple candidate workflows created, the KPI change vector from the existing workflow is determined. Multiple combined measures are generated by combining two or more of the KPI fluctuation vectors from among the multiple KPI fluctuation vectors. If a single selection route in the candidate workflow includes multiple changes, the weight parameters are set such that the weight of the higher-level changes in the selection route is higher than the weight of the lower-level changes in the selection route, and the predicted KPI values ​​for each of the multiple composite measures are calculated accordingly. Output the workflow of the composite measures that satisfy the KPI forecast value. A workflow generation method characterized by having a computer perform the processing.

2. The process for calculating the predicted KPI values ​​for each of the aforementioned multiple composite measures is to calculate the predicted KPI values ​​by weighted summing the predicted KPI values ​​of the existing workflow and the KPI fluctuation vector. A workflow generation method according to claim 1, characterized by including processing.

3. The weight parameter is set according to the flow depth of the candidate workflow. A workflow generation method according to claim 1 or 2, characterized in that...

4. We created several candidate workflows by modifying parts of an existing workflow that defines multiple conditional branches and the follow-up actions to be taken at each branch. For each of the multiple candidate workflows created, the KPI change vector from the existing workflow is determined. Multiple combined measures are generated by combining two or more of the KPI fluctuation vectors from among the multiple KPI fluctuation vectors. If a single selection route in the candidate workflow includes multiple changes, the weight parameters are set such that the weight of the higher-level changes in the selection route is higher than the weight of the lower-level changes in the selection route, and the predicted KPI values ​​for each of the multiple composite measures are calculated accordingly. Output the workflow of the composite measures that satisfy the KPI forecast value. A workflow generation program characterized by having a computer execute the processing.

Images