Search program, search device, and search method

Abstract

To retrieve an appropriate case about an intervention method to an intervention object person.SOLUTION: A computer selects a specific intervention method from among a plurality of intervention methods on the basis of first characteristic information showing a characteristic of an intervention object person. The computer calculates an estimated characteristic change amount representing difference between second characteristic information showing a characteristic after applying the specific intervention method to the intervention object person and the first characteristic information. The computer retrieves specific case data from case data including the intervention method, character information and characteristic change amount of each of a plurality of persons. The specific case data includes a combination of characteristic information and a characteristic change amount similar to the combination of the first characteristic information, and the estimated characteristic change amount, and the specific intervention method. The computer outputs information based on the specific case data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] This invention relates to a search technology. [Background technology]

[0002] National and local governments are implementing initiatives to reduce social insurance premiums by extending the healthy life expectancy of residents, and an increasing number of local governments are providing outreach support. In outreach support, public health nurses from local governments visit the homes of elderly people and engage with them. Through this engagement, it is possible to increase the elderly's interest in frailty prevention, change their attitudes, and encourage them to take actions that lead to preventative care. This engagement is sometimes called intervention.

[0003] In relation to health improvement, information generation systems are known that can generate information to verify the health-improving effects of products or services while reducing the burden of time, effort, and cost associated with verification (see, for example, Patent Document 1). Information management systems are also known that can collect more information related to user health (see, for example, Patent Document 2).

[0004] Behavioral support systems are known that can improve the motivation and self-efficacy of individuals regarding their actions and suppress the occurrence of relapses (see, for example, Patent Document 3). Information processing devices are also known that are easy for users to use when creating care plans and assist in creating care plans that are more appropriate for the person receiving care (see, for example, Patent Document 4). [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2020-87349 [Patent Document 2] Japanese Patent Publication No. 2020-27309 [Patent Document 3] Japanese Patent Publication No. 2016-85703 [Patent Document 4] Japanese Patent Publication No. 2019-46474 [Overview of the project] [Problems that the invention aims to solve]

[0006] In outreach support, it is desirable to determine the approach to reach out to the target group by referring to past examples of outreach efforts made to the elderly. However, it is not easy to extract the most effective successful examples from a large number of cases.

[0007] Furthermore, this problem arises not only when deciding on methods of engaging with the elderly in outreach support, but also when deciding on methods of intervention for target groups in various social activities.

[0008] In one aspect, the present invention aims to find appropriate examples of intervention methods for individuals subject to intervention. [Means for solving the problem]

[0009] One approach involves having the search program perform the following actions on the computer:

[0010] The computer selects a specific intervention method from several options based on first characteristic information that describes the characteristics of the person to be intervened. The computer then calculates an estimated change in characteristics, which represents the difference between the first characteristic information and second characteristic information that describes the characteristics of the person after the specific intervention method has been applied to them.

[0011] The computer searches for specific case data from a dataset containing intervention methods, characteristic information, and characteristic change amounts for each of multiple individuals. The specific case data includes combinations of characteristic information and characteristic change amounts similar to the first characteristic information and estimated characteristic change amount combination, along with a specific intervention method. The computer outputs information based on the specific case data. [Effects of the Invention]

[0012] According to one aspect, appropriate cases regarding an intervention method for an intervention target person can be retrieved.

Brief Description of the Drawings

[0013] [Figure 1] It is a functional configuration diagram of the retrieval device according to the embodiment. [Figure 2] It is a flowchart of the first retrieval process. [Figure 3] It is a functional configuration diagram showing a specific example of the retrieval device. [Figure 4] It is a diagram showing a factor score matrix. [Figure 5] It is a diagram showing a state matrix indicating the state before the application. [Figure 6] It is a diagram showing a state estimation model. [Figure 7] It is a diagram showing an application matrix. [Figure 8] It is a diagram showing a state matrix indicating the state after the application. [Figure 9] It is a diagram showing a factor score estimation model. [Figure 10] It is a diagram showing a combination model. [Figure 11] It is a diagram showing case data. [Figure 12] It is a diagram showing a factor score vector. [Figure 13] It is a diagram showing state information generated from the factor score vector before the application. [Figure 14] It is a diagram showing state information generated from the factor score vector after the application. [Figure 15] It is a diagram showing situation information. [Figure 16A] It is a flowchart of the learning process (Part 1). [[ID=S8]] [Figure 16B] It is a flowchart of the learning process (Part 2). [Figure 17A] It is a flowchart of the second retrieval process (Part 1). [Figure 17B] This is the flowchart (part 2) for the second search process. [Figure 18] This is a hardware configuration diagram of an information processing device. [Modes for carrying out the invention]

[0014] The embodiments will be described in detail below with reference to the drawings.

[0015] For example, when a public health nurse visits the homes of elderly people to provide outreach support, the following information is collected:

[0016] (A1) Questionnaire results regarding the residents' condition, such as a basic checklist or assessment sheet. (A2) Classification labels indicating the methods of outreach to residents. (A3) Changes in residents' awareness or interest (for example, the state of residents before and after the intervention)

[0017] The results of the questionnaire in (A1) are collected before intervention, but not after. The basic checklist includes multiple questions, such as those on daily life, exercise, nutrition, oral health, social withdrawal, cognition, and depression. Methods of intervention in (A2) include providing information, feedback, modeling, and guidance. Information indicating the stage of the behavioral change model is used to show the state of residents before and after intervention in (A3).

[0018] By associating the information in (A1) to (A3) with situational information describing the circumstances at the time of the visit, case data is generated and stored in the database. Examples of situational information include the following:

[0019] (B1) Public health nurse who took action (B2) Information on specific actions rather than classification labels (B3) Types of pamphlets distributed to residents (B4) Key points for deciding on the approach (B5) Action goals agreed upon during the visit (B6) Attendees (B7) Other (things that caught your attention, etc.)

[0020] When determining how to approach target individuals, it is desirable to search the database for case data containing the most effective approach and present it to the public health nurse. Therefore, we will describe a search device for comparative examples that retrieves case data from the database. The search device for comparative examples searches for case data using the following procedure.

[0021] (1) The search device generates a consciousness change estimation model by training a deep neural network (DNN) using machine learning with training data. The training data includes survey results and intervention methods as input data, and the results of consciousness change are included as training data for the output data. Intervention methods are represented, for example, by one-hot vectors. As for the results of consciousness change, three types of results are used, for example, change in the positive direction, no change, or change in the negative direction (counterproductive effect).

[0022] (2) The search device uses an awareness change estimation model to generate results for awareness change from the combination of questionnaire results obtained from the target individuals and each of the multiple intervention methods. The search device then selects the intervention method with the highest probability of showing a positive change as the recommended intervention method.

[0023] (3) The search device searches the database for case data that includes classification labels indicating recommended intervention methods and also includes questionnaire results similar to those of the target individuals, and presents them to the public health nurse. Questionnaire results similar to those of the target individuals mean questionnaire results with a small difference from the target individuals' questionnaire results.

[0024] Past case data that includes recommended intervention methods and shows similar survey results are valuable case data that can be used as a reference when determining the specific content of intervention methods for target individuals. Public health nurses can refer to the specific intervention content described in the presented case data to decide whether or not to adopt the recommended intervention methods for target individuals.

[0025] However, just because survey results are similar does not necessarily mean that the same intervention methods will be effective for the target group. For example, if a basic checklist includes multiple questions across a wide range of categories, the effectiveness of the intervention method will differ depending on which questions the answers are the same and which questions the answers are different.

[0026] As an example, consider a scenario where the survey results of the target group show "yes" to a specific question, and the answer to that question is important for determining the effectiveness of the intervention.

[0027] In this case, case data from successful interventions, where the answer to that question is "no" and the answers to all other questions are consistent, may be extracted as case data containing similar survey results. Even if the extracted case data corresponds to a successful case, there is a risk that this case data may not lead to successful interventions even if public health nurses refer to it.

[0028] Figure 1 shows an example of the functional configuration of a search device according to an embodiment. The search device 101 in Figure 1 includes a selection unit 111, an estimation unit 112, a search unit 113, and an output unit 114.

[0029] Figure 2 is a flowchart showing an example of the first search process performed by the search device 101 in Figure 1. First, the selection unit 111 selects a specific intervention method from among multiple intervention methods based on first characteristic information indicating the characteristics of the person to be intervened (step 201). Then, the estimation unit 112 calculates the estimated characteristic change amount, which represents the difference between the first characteristic information and second characteristic information indicating the characteristics after the specific intervention method has been applied to the person to be intervened (step 202).

[0030] Next, the search unit 113 searches for specific case data from case data that includes intervention methods, characteristic information, and characteristic change amounts for each of multiple individuals (step 203). The specific case data includes combinations of characteristic information and characteristic change amounts similar to the combination of first characteristic information and estimated characteristic change amount, as well as a specific intervention method. The output unit 114 outputs information based on the specific case data (step 204).

[0031] According to the search device 101 in Figure 1, it is possible to search for appropriate cases regarding intervention methods for individuals subject to intervention.

[0032] Figure 3 shows a specific example of the search device 101 in Figure 1. The search device 301 in Figure 3 includes an analysis unit 311, a training unit 312, a generation unit 313, a selection unit 314, an estimation unit 315, a search unit 316, a display unit 317, and a storage unit 318. The selection unit 314, estimation unit 315, search unit 316, and display unit 317 correspond to the selection unit 111, estimation unit 112, search unit 113, and output unit 114 in Figure 1, respectively.

[0033] The search device 301 assists the user in determining how to approach the target person. The target person corresponds to the person to be intervened, and the approach method corresponds to the intervention method. For example, in the case of outreach support, the user corresponds to the public health nurse who will visit the target person's home to conduct the intervention. The search device 301 performs learning processing and search processing.

[0034] During the learning process, the memory unit 318 stores questionnaire data 321 from N individuals who have been previously interacted with. N is an integer greater than or equal to 2, but the number of people required for factor analysis and model estimation, described later, is set to N. The questionnaire data 321 includes questionnaire results, such as basic checklists and / or assessment sheets, obtained from each individual before the interaction.

[0035] The survey results include responses to each of the K questions (where K is an integer greater than or equal to 2). The K questions are an example of multiple items. The responses to each question are represented, for example, by numerical values, and the survey data 321 is represented by an N x K matrix Z, which contains, for example, the K responses for each person as K-dimensional vectors.

[0036] The analysis unit 311 performs exploratory factor analysis on the questionnaire data 321 to assign meaningful weights to the questionnaire results. First, the analysis unit 311 performs exploratory factor analysis to obtain M principal factors and a K x M factor loading matrix A. Each element of the factor loading matrix A represents a factor loading.

[0037] Next, the analysis unit 311 obtains a K x K correlation matrix R from the questionnaire data 321, and the inverse matrix R of the correlation matrix R -1 The analysis unit 311 then calculates the coefficient matrix C, which is a K x M matrix that transforms matrix Z into an N x M factor score matrix F, using the following formula, and stores the coefficient information 322 representing the coefficient matrix C in the storage unit 318.

[0038] C=R -1 A (1)

[0039] Next, the analysis unit 311 obtains the factor score matrix F using the following formula.

[0040] F=ZC (2)

[0041] Each element of the factor score matrix F represents a factor score. The factor score is a weighted sum of each person's K responses, representing each person's evaluation value for each principal factor. Each element of the coefficient matrix C represents the weight of each response in the calculation of the factor score for each principal factor of each person. Each row of the factor score matrix F corresponds to characteristic information that indicates the characteristics of the person.

[0042] By calculating factor scores, which are the weighted sum of K responses, we can quantify the importance of potential principal factors that influence the survey results.

[0043] Figure 4 shows an example of a factor score matrix F. Pi (i=1~N) is the identifier of the i-th person's questionnaire result, and Rj (j=1~M) is the identifier of the j-th principal factor. In this example, M=5, R1 represents self-efficacy, R2 represents motivation, R3 represents physical ability, R4 represents social support, and R5 represents other principal factors.

[0044] The analysis unit 311 may obtain the coefficient matrix C using a confirmatory factor analysis, which presupposes the connections between each question and each factor, instead of exploratory factor analysis.

[0045] Next, the training unit 312 generates training data 323 and stores it in the storage unit 318. The training data 323 includes a factor score matrix F and state information corresponding to each row of the factor score matrix F. The state information is training data that shows the state of each person before intervention, corresponding to each row of the factor score matrix F, and is prepared in advance along with the questionnaire data 321. The state information of N people is represented by an N x L state matrix, where L represents the number of states.

[0046] Figure 5 shows an example of a state matrix representing the state before intervention. Tj (j=1~L) is the identification information for the j-th state. In this example, L=5, and the stages in the Stages of Behavior Model are used as states. T1 represents the precontemplation stage, T2 represents the contemplation stage, T3 represents the preparation stage, T4 represents the action stage, and T5 represents the maintenance stage. As j increases, Tj represents a more advanced stage.

[0047] Each row in the state matrix represents state information indicating the state of a person before any interaction, and is represented by a one-hot vector. For example, the state information of a person corresponding to survey result P1 is (0,1,0,0,0), which corresponds to T2.

[0048] Next, the training unit 312 generates a state estimation model 325 by training a machine learning model using training data 323, and stores it in the memory unit 318. The state estimation model 325 estimates the state of a person from a factor score vector containing M factor scores of that person. A model generated by machine learning is sometimes called a trained model.

[0049] Machine learning models such as neural networks, random forests, or support vector machines can be used. Neural networks may also be DNNs. Machine learning algorithms such as backpropagation can be used.

[0050] Figure 6 shows an example of state estimation model 325. The state estimation model 601 in Figure 6 is a DNN, with the input layer 611 containing nodes 621-1 to 621-5, and the output layer 612 containing nodes 622-1 to 622-5. Nodes 621-1 to 621-5 are input nodes that accept factor scores corresponding to R1 to R5, respectively, and nodes 622-1 to 622-5 are output nodes that output numerical values ​​of the state corresponding to T1 to T5, respectively.

[0051] Next, the training unit 312 generates training data 324 and stores it in the memory unit 318. The training data 324 includes a factor score matrix F, questionnaire data 321, the methods of intervention performed on each person, and state information indicating the state of each person after the intervention. The intervention methods and state information are prepared in advance along with the questionnaire data 321.

[0052] The methods of intervention applied to each individual are examples of intervention methods corresponding to each of the multiple trait information items. The state information indicating the state of each individual after intervention is an example of state information corresponding to the combination of each of the multiple trait information items and the intervention method corresponding to each of the multiple trait information items.

[0053] The ways in which N people interact with each other are represented by an N x G interaction matrix, where G represents the number of interaction methods.

[0054] Figure 7 shows an example of an intervention matrix. Hj (j=1~G) is the identification information for the j-th intervention method. In this example, G=5. As identification information for intervention methods, for example, classification labels indicating the intervention method can be used. H1 represents information provision, H2 represents feedback, H3 represents modeling, H4 represents instruction, and H5 represents other intervention methods.

[0055] Each row in the interaction matrix is ​​a one-hot vector representing the interaction method. For example, the interaction method for the person corresponding to survey result P1 is (1,0,0,0,0), which corresponds to H1.

[0056] Figure 8 shows an example of a state matrix indicating the state after an intervention. For example, the state information of a person corresponding to questionnaire result P1 is (0,0,1,0,0), which corresponds to T3. Comparing Figure 5 and Figure 8, we can see that the intervention in Figure 7 changed this person's state from the contemplation stage to the preparation stage, indicating that the intervention was successful.

[0057] Next, the training unit 312 generates a factor score estimation model 326 by training a machine learning model using the state estimation model 325 and training data 324, and stores it in the memory unit 318. The factor score estimation model 326 is a trained model that estimates the factor score vector of a person after intervention, based on the person's factor score vector before intervention, questionnaire results, and intervention method. The factor score estimation model 326 is an example of a characteristic estimation model.

[0058] Figure 9 shows an example of a factor score estimation model 326. The factor score estimation model 901 in Figure 9 is a DNN, with input layer 911 containing nodes 921-1 to 921-5, input layer 912 containing nodes 922-1 to 922-K, and input layer 913 containing nodes 923-1 to 923-5. Output layer 914 contains nodes 924-1 to 924-5.

[0059] Nodes 921-1 to 921-5 are input nodes that accept pre-intervention factor scores corresponding to R1 to R5, respectively. Nodes 922-1 to 922-K are input nodes that accept K responses included in the questionnaire results. Qj (j=1 to K) is the identification information for the j-th question. Nodes 923-1 to 923-5 are input nodes that accept one-hot vectors indicating one of H1 to H5.

[0060] Nodes 924-1 to 924-5 are output nodes that output the factor scores after intervention, corresponding to R1 to R5, respectively.

[0061] Multiple hidden layers perform calculations on the numerical values ​​input to input layers 912 and 913, and the calculation result of the final hidden layer is output to output layer 914. On the other hand, the factor score vector input to input layer 911 before any intervention is output directly to output layer 914 via the identity map.

[0062] The output layer 914 outputs the sum of a vector representing the weighted sum of the calculation results of the final hidden layer and the factor score vector input to the input layer 911 as the factor score vector after the intervention. In this case, the factor score vector before the intervention is used as the bias in the output layer 914.

[0063] To simplify the explanation, let's assume that the factor score estimation model 901 contains only one hidden layer. In this case, the factor score vector y after the intervention is expressed by the following equation, using a vector x whose elements are the numerical values ​​input to input layers 912 and 913, a bias vector b1, and the factor score vector b2 before the intervention.

[0064] y = W2·f(W1·x+b1)+b2 (3)

[0065] W1 represents the weight for vector x, f represents the activation function, and W2 represents the weight for f(W1·x+b1). The W2·f(W1·x+b1) on the right side of equation (3) represents the difference between the factor score vector y and the factor score vector b2.

[0066] Figure 10 shows an example of a combinatorial model in machine learning that generates a factor score estimation model 326. The combinatorial model 1001 in Figure 10 includes a DNN 1011 corresponding to the machine learning model before training, and a state estimation model 601 in Figure 6. The DNN 1011 has a similar configuration to the factor score estimation model 901 in Figure 9. The input layer 611 of the state estimation model 601 is connected to the output layer 914 of the DNN 1011.

[0067] The training unit 312 provides the input layers 911 to 913 of the DNN1011 with the factor score vectors before intervention corresponding to each row of the factor score matrix F, the K-dimensional vector representing each person's response, and the one-hot vectors corresponding to each row of the intervention matrix in Figure 7. The training unit 312 then provides the state information corresponding to each row of the state matrix in Figure 8 as training data to the output layer 612 of the state estimation model 601, and trains the DNN1011 by backpropagation learning. The factor score vectors before intervention input to the input layer 911 are used as initial values ​​for the bias in the output layer 914.

[0068] In backpropagation learning, the training unit 312 sets the learning rate of the bias in the output layer 914 and the learning rate of each layer of the state estimation model 601 to 0. Therefore, the bias applied to the output layer 914 is not updated, and the state estimation model 601 is not updated. Through backpropagation learning using this combinatorial model 1001, the factor score estimation model 901 is generated from the DNN 1011.

[0069] The factor score estimation model 326 may include an input layer that accepts a factor score vector before intervention, instead of an input layer that accepts K responses included in the questionnaire results.

[0070] Next, the training unit 312 generates N case data corresponding to the questionnaire results of N people included in the questionnaire data 321, and stores these case data as a case set 327 in the storage unit 318. The case set 327 is used as a database for accumulating case data.

[0071] Figure 11 shows an example of case data included in case set 327. The case data in Figure 11 includes entry number, intervention method, factor score vector before intervention, difference vector, state information before intervention, state information after intervention, and situation information.

[0072] The entry number is the identification information for the case data. The intervention method represents the method of intervention performed on the person. The factor score vector before intervention is a vector consisting of M factor scores contained in the row corresponding to that person in the factor score matrix F.

[0073] The difference vector is a vector that represents the difference between the factor score vector after the intervention and the factor score vector before the intervention.

[0074] The training unit 312 inputs each person's factor score vector before intervention, questionnaire results, and intervention method into the factor score estimation model 326, and obtains the factor score vector output from the factor score estimation model 326 as the person's factor score vector after intervention. The training unit 312 then calculates the difference between each element in the factor score vector after intervention and the same element in the factor score vector before intervention, and generates a difference vector containing M differences as elements. The difference vector is an example of a characteristic change.

[0075] Pre-intervention state information indicates the state of the person before the intervention. Post-intervention state information indicates the state of the person after the intervention. Situation information indicates the situation when the intervention occurred.

[0076] For example, the information described in (B1) to (B7) above can be used as situational information. The focus point in (B4) may be, for example, a specific question included in the survey results, and the behavioral goal in (B5) may be, for example, doing exercises at home. The person present in (B6) may be, for example, a family member who was present during the visit, and the thing that caught their attention in (B7) may be, for example, the personality of the person who received the intervention. The person's personality may be one of logical thinking.

[0077] The search device 301 may add case data of other cases different from the N cases corresponding to the questionnaire data 321 to the case set 327.

[0078] During the retrieval process, the storage unit 318 stores the questionnaire data 328 of the new target person. The questionnaire data 328 includes the same questionnaire results as the questionnaire data 321. The questionnaire data 328 is an example of information obtained from the target person for each of the multiple items.

[0079] The generation unit 313 generates a factor score vector 329 of the target person before intervention based on the questionnaire data 328 and stores it in the storage unit 318. The generation unit 313 uses a vector u containing the K numerical values ​​included in the questionnaire data 328 as elements and a coefficient matrix C shown by the coefficient information 322 to obtain the factor score vector v using the following formula.

[0080] v=uC (4)

[0081] The generation unit 313 then stores the factor score vector v as a factor score vector 329 in the storage unit 318. By generating the factor score vector 329 from the questionnaire data 328, the responses of the target individuals to the K questions can be converted into evaluation values ​​for each principal factor of the target individuals.

[0082] Factor score vector 329 is an example of primary characteristic information that shows the characteristics of the intervention subjects. The M factor scores included in factor score vector 329 are examples of the numerical values ​​of multiple factors related to multiple items.

[0083] Figure 12 shows an example of factor score vector 329. Factor score vector 329 in Figure 12 contains the factor scores for R1 to R5. In this example, M=5.

[0084] Next, the selection unit 314 selects a specific intervention method suitable for the target person from among the G intervention methods based on the factor score vector 329, and uses that as the recommended intervention method. Then, the selection unit 314 stores the intervention information 330 indicating the recommended intervention method in the storage unit 318.

[0085] The selection unit 314 inputs the factor score vector 329 into the state estimation model 325, obtains L numerical values ​​for states output from the state estimation model 325, and estimates the state of the target person before intervention based on these numerical values. For example, the selection unit 314 estimates that the state corresponding to the maximum value among the L numerical values ​​is the state of the target person before intervention. The L numerical values ​​output from the state estimation model 325 are an example of first state information.

[0086] By using the state estimation model 325, it is possible to accurately estimate the state before an intervention.

[0087] Figure 13 shows an example of state information generated from the factor score vector 329 in Figure 12. The state information in Figure 13 includes numerical values ​​for each state from T1 to T5. In this example, L=5. In this case, since the value for T1 is the largest, the state of the person being addressed before intervention is estimated to be the precontemplation stage.

[0088] Next, the selection unit 314 sequentially selects each intervention method from the G intervention methods and uses the factor score estimation model 326 to estimate the factor score vector of the target person after intervention when the selected intervention method is applied.

[0089] The selection unit 314 inputs the factor score vector 329 before intervention, the questionnaire data 328, and the selected intervention method into the factor score estimation model 326. The selection unit 314 then obtains the factor score vector output from the factor score estimation model 326 as the factor score vector of the intervention target. The factor score vector of the intervention target is an example of third-characteristic information that shows the characteristics after each of the multiple intervention methods has been applied to the intervention target.

[0090] Next, the selection unit 314 inputs the factor score vector of the person being worked on after the intervention into the state estimation model 325, obtains L numerical values ​​of states output from the state estimation model 325, and estimates the state of the person being worked on after the intervention based on these numerical values. For example, the selection unit 314 estimates that the state corresponding to the maximum value among the L numerical values ​​is the state of the person being worked on after the intervention. The L numerical values ​​output from the state estimation model 325 are an example of second state information.

[0091] By using the factor score estimation model 326 and the state estimation model 325, the state after an intervention can be estimated with high accuracy.

[0092] Figure 14 shows an example of state information generated from the factor score vector after intervention for the target individual. In this example, H2 feedback is selected as the intervention method. In this case, since the T2 value is the maximum, the target individual's state after intervention is estimated to be the contemplation stage.

[0093] Next, the selection unit 314 compares the state after each intervention method is applied and selects the intervention method that best improves the state as the recommended intervention method. This makes it possible to identify the intervention method with the highest probability of success as the recommended intervention method.

[0094] In the Stages of Behavior Model, a stage further than the stage before intervention corresponds to an improved state. For example, stages T2 to T5 correspond to a stage of improvement beyond the stage T1.

[0095] If the state after intervention using feedback is T2, and the state after intervention using other intervention methods is T1, then feedback is selected as the recommended intervention method. If the state after intervention using both feedback and modeling is T2, and the state after intervention using other intervention methods is T1, then the intervention method with the larger T2 value among feedback and modeling is selected as the recommended intervention method. Furthermore, if the state after intervention using multiple intervention methods is T2, then the intervention method with the largest T2 value shown in Figure 14 is selected as the recommended intervention method.

[0096] Next, the estimation unit 315 inputs the factor score vector 329, the questionnaire data 328, and the recommended intervention method indicated by the intervention information 330 into the factor score estimation model 326. Then, the estimation unit 315 obtains the factor score vector 331 after intervention, which is output from the factor score estimation model 326, and stores it in the storage unit 318. The factor score vector 331 is an example of second characteristic information that shows the characteristics of the intervention target after a specific intervention method has been applied.

[0097] Next, the estimation unit 315 calculates the difference between each element in the factor score vector 331 and the same element in the factor score vector 329, and generates an estimated difference vector 332 containing M differences as elements. Then, the estimation unit 315 stores the estimated difference vector 332 in the storage unit 318. The estimated difference vector 332 is an example of an estimated characteristic change amount that represents the difference between the second characteristic information and the first characteristic information.

[0098] By using the factor score estimation model 326, the factor score vector 331 after the intervention can be estimated with high accuracy. Consequently, the accuracy of the estimated difference vector 332 generated from the factor score vector 331 is also improved.

[0099] Next, the search unit 316 searches for specific case data suitable for the target person from the case set 327. The specific case data includes the same approach method as the recommended approach method selected by the selection unit 314, and includes a combination C2 of a pre-approach factor score vector and a difference vector similar to the combination C1 of the factor score vector 329 and the estimated difference vector 332.

[0100] The search unit 316 extracts case data including the same approach method as the recommended approach method from the case set 327, and calculates the distance D between the combination C1 and the combination C2 included in the extracted case data by the following formula.

[0101] D = α(v - s)(Σ1) -1 (v - s) T + (1 - α)(Δv - Δs)(Σ2) -1 (Δv - Δs) T (5)

[0102] v represents the factor score vector 329, s represents the pre-approach factor score vector included in the extracted case data. Δv represents the estimated difference vector 332, and Δs represents the difference vector included in the extracted case data.

[0103] The vectors v, vector s, vector Δv, and vector Δs are row vectors. Therefore, (v - s) and (Δv - Δs) are also row vectors. (v - s) T represents a column vector obtained by transposing (v - s), and (Δv - Δs) T represents a column vector obtained by transposing (Δv - Δs).

[0104] Σ1 represents the covariance matrix of the pre-approach factor score vectors included in the case data within the case set 327, and (Σ1) -1 represents the inverse matrix of Σ1. Σ2 represents the covariance matrix of the difference vectors included in the case data within the case set 327, and (Σ2) -1 represents the inverse matrix of Σ2.

[0105] α is a real number in the range of 0 to 1. The first term on the right-hand side of equation (5) represents the product of the Mahalanobis distance between vectors v and s and α. The second term represents the product of the Mahalanobis distance between vectors Δv and Δs and (1-α). By changing α, you can adjust which of the first or second term is given more weight.

[0106] The similarity between combination C1 and combination C2 increases as the distance D decreases and decreases as the distance D increases. Therefore, the search unit 316 determines the specific case data to be the case data with the smallest distance D among the extracted case data.

[0107] By calculating the distance D using equation (5), it is possible to search for appropriate success stories where not only are the factor score vectors before intervention similar, but the amount of change in the factor score vectors due to the intervention is also similar. Therefore, by considering not only the state of the target person before intervention, but also the similarity of the effect (change) after the intervention, it is possible to obtain information on successful stories that are most useful in determining the method of intervention for the target person.

[0108] The distance between vector v and vector s, and the distance between vector Δv and vector Δs, may be any inter-vector distance other than the distance shown in equation (5). The inter-vector distance may be Euclidean distance or Manhattan distance. The search unit 316 may use inter-vector similarity instead of inter-vector distance to obtain specific case data.

[0109] Next, the search unit 316 generates search results 333 using specific case data and stores them in the storage unit 318. The search results 333 include, for example, the intervention method included in the specific case data, the state indicated by the state information before the intervention, the state indicated by the state information after the intervention, and situation information. The display unit 317 displays the search results 333 on the screen. The search results 333 are an example of information based on specific case data.

[0110] Figure 15 shows an example of situational information included in search result 333. The person in charge corresponds to information (B1), the pamphlet corresponds to information (B3), and the focus points correspond to information (B4). The agreed-upon action goals correspond to information (B5), the attendees correspond to information (B6), and others correspond to information (B7).

[0111] By displaying situational information along with the intervention methods used in past cases, public health nurses can refer to the circumstances under which those intervention methods were applied and decide whether or not to adopt the same intervention method.

[0112] The search device 301 in Figure 3 can be used not only for outreach support but also in various social activities to determine intervention methods for target individuals. For example, the target individuals may be employees of a company, and the intervention may be aimed at improving employee performance.

[0113] Figures 16A and 16B are flowcharts illustrating examples of the learning process performed by the search device 301 in Figure 3. First, the analysis unit 311 determines the number of principal factors M in the exploratory factor analysis (step 1601). The analysis unit 311 determines the number of principal factors M, for example, according to user instructions. The number of principal factors M is determined using existing methods, such as selecting the number of factors whose eigenvalues ​​in the correlation matrix are 1 or greater.

[0114] Next, the analysis unit 311 obtains the factor loading matrix A by performing exploratory factor analysis on the questionnaire data 321 of N people (step 1602).

[0115] Next, the analysis unit 311 calculates the correlation matrix R from the questionnaire data 321 (step 1603) and checks whether the calculation has converged (step 1604). If the correlation matrix is ​​singular and the calculation does not converge (step 1604, NO), the search device 301 terminates the process.

[0116] If the calculation converges (Step 1604, YES), the analysis unit 311 calculates the inverse matrix R of the correlation matrix R. -1 The first unit calculates the coefficient matrix C using equation (1) and generates coefficient information 322 representing the coefficient matrix C (step 1605). Then, the analysis unit 311 calculates the factor score matrix F using equation (2) (step 1606).

[0117] Next, the training unit 312 performs training to generate a state estimation model 325 using machine learning with the training data 323 (step 1607), and checks whether the machine learning has converged (step 1608). If the machine learning has not converged (step 1608, NO), the training unit 312 repeats the process from step 1607 onward.

[0118] If the machine learning model converges (Step 1608, YES), the training unit 312 generates a combinatorial model by connecting the pre-training machine learning model and the state estimation model 325 (Step 1609). Then, the training unit 312 performs training to generate a factor score estimation model 326 using machine learning with the training data 324 (Step 1610), and checks whether the machine learning model has converged (Step 1611). If the machine learning model has not converged (Step 1611, NO), the training unit 312 repeats the process from Step 1610 onward.

[0119] If the machine learning converges (Step 1611, YES), the training unit 312 selects one survey result from the survey data 321 (Step 1612). Then, the training unit 312 inputs the selected survey result, the pre-intervention factor score vector corresponding to that survey result, and the intervention method into the factor score estimation model 326 to generate the post-intervention factor score vector (Step 1613).

[0120] Next, the training unit 312 generates a difference vector from the factor score vector after intervention and the factor score vector before intervention (step 1614). Then, the training unit 312 adds case data, including entry number, intervention method, factor score vector before intervention, difference vector, state information before intervention, state information after intervention, and situation information, to the case set 327 (step 1615).

[0121] Next, the training unit 312 checks whether all the survey results included in the survey data 321 have been selected (step 1616). If there are any unselected survey results remaining (step 1616, NO), the training unit 312 repeats the processing from step 1612 onward for the next survey result. If all survey results have been selected (step 1616, YES), the search device 301 terminates the process.

[0122] Figures 17A and 17B are flowcharts illustrating an example of the second search process performed by the search device 301 in Figure 3. First, the generation unit 313 uses a vector u containing K numerical values ​​from the questionnaire data 328 of the target individuals as elements, and the coefficient matrix C shown by the coefficient information 322, to obtain a factor score vector v using equation (4) (step 1701). Then, the generation unit 313 stores the factor score vector v as a factor score vector 329 in the storage unit 318.

[0123] Next, the selection unit 314 inputs the factor score vector 329 into the state estimation model 325 (step 1702), and estimates the state of the target person before intervention based on the L numerical values ​​output from the state estimation model 325 (step 1703).

[0124] Next, the selection unit 314 inputs the factor score vector 329 and the questionnaire data 328 into the factor score estimation model 326 (step 1704). Then, the selection unit 314 selects one intervention method from among the G intervention methods (step 1705) and inputs the selected intervention method into the factor score estimation model 326 (step 1706). Finally, the selection unit 314 obtains the factor score vector output from the factor score estimation model 326 as the factor score vector of the intervention target.

[0125] Next, the selection unit 314 inputs the factor score vector of the person being worked on after the intervention into the state estimation model 325, estimates the state of the person being worked on after the intervention based on the L numerical values ​​output from the state estimation model 325, and records it in the storage unit 318 (step 1707).

[0126] Next, the selection unit 314 checks whether all intervention methods have been selected (step 1708). If there are still unselected intervention methods (step 1708, NO), the selection unit 314 repeats the process from step 1705 onwards for the next intervention method.

[0127] If all intervention methods are selected (step 1708, YES), the selection unit 314 compares the post-intervention state for each intervention method recorded in the storage unit 318 and selects the intervention method that best improves the state as the recommended intervention method (step 1709). The selection unit 314 then generates intervention information 330 indicating the recommended intervention method.

[0128] Next, the estimation unit 315 inputs the factor score vector 329, the questionnaire data 328, and the recommended intervention method indicated by the intervention information 330 into the factor score estimation model 326 and estimates the factor score vector 331 after the intervention (step 1710). Then, the estimation unit 315 generates an estimated difference vector 332 from the factor score vector 331 and the factor score vector 329 (step 1711).

[0129] Next, the search unit 316 extracts case data including recommended intervention methods from the case set 327 (step 1712) and sets an initial value for the minimum value DM of distance D (step 1713). The initial value of the minimum value DM is set to be a value that is sufficiently larger than the expected maximum value of distance D.

[0130] Next, the search unit 316 selects one case data from the extracted case data (step 1714). Then, the search unit 316 calculates the distance D using equation (5) with the factor score vector 329, the estimated difference vector 332, and the factor score vector and difference vector included in the selected case data (step 1715).

[0131] Next, the search unit 316 compares the distance D with the minimum value DM (step 1716). If the distance D is less than the minimum value DM (step 1716, YES), the search unit 316 sets the distance D to the minimum value DM and records the entry number of the selected case data in the storage unit 318 (step 1717).

[0132] Next, the search unit 316 checks whether all extracted case data has been selected (step 1718). If there is still unselected case data remaining (step 1718, NO), the search unit 316 repeats the processing from step 1714 onwards for the next case data.

[0133] If all case data is selected (step 1718, YES), the search unit 316 generates search results 333 using the specific case data indicated by the last recorded entry number (step 1719). Then, the display unit 317 displays the search results 333 on the screen (step 1720).

[0134] If the distance D is greater than or equal to the minimum value DM (step 1716, NO), the search unit 316 performs the processing from step 1718 onwards.

[0135] The configurations of the search device 101 in Figure 1 and the search device 301 in Figure 3 are merely examples, and some components may be omitted or changed depending on the application or conditions of the search device 101 or the search device 301. For example, if the coefficient information 322, state estimation model 325, factor score estimation model 326, and case set 327 in Figure 3 are generated by an external device, the analysis unit 311 and the training unit 312 can be omitted.

[0136] The flowcharts in Figures 2, 16A, 16B, 17A, and 17B are merely examples, and some processes may be omitted or modified depending on the configuration or conditions of the search device 101 or search device 301. For example, if the coefficient information 322, state estimation model 325, factor score estimation model 326, and case set 327 in Figure 3 are generated by an external device, the learning processes in Figures 16A and 16B can be omitted.

[0137] The factor score matrix shown in Figure 4, the state matrices shown in Figures 5 and 8, and the interaction matrix shown in Figure 7 are merely examples, and these matrices will change depending on the case in which the questionnaire data 321 was obtained. The state estimation model 601 shown in Figure 6, the factor score estimation model 901 shown in Figure 9, and the combination model 1001 shown in Figure 10 are merely examples, and other machine learning models may be used. The case data shown in Figure 11 is merely an example, and case data in a different format may be used.

[0138] The factor score vector 329 shown in Figure 12 and the state information shown in Figures 13 and 14 are merely examples; the factor score vector 329 and state information change according to the questionnaire data 328. The situation information shown in Figure 15 is also merely an example; the situation information changes according to the factor score vector 331 after intervention.

[0139] Equations (1) to (5) are merely examples, and the search device 301 may perform learning and search processing using other calculation formulas.

[0140] Figure 18 shows an example of the hardware configuration of an information processing device used as the search device 101 in Figure 1 and the search device 301 in Figure 3. The information processing device in Figure 18 includes a CPU (Central Processing Unit) 1801, memory 1802, input device 1803, output device 1804, auxiliary storage device 1805, media drive device 1806, and network connection device 1807. These components are hardware and are connected to each other by a bus 1808.

[0141] Memory 1802 is, for example, a semiconductor memory such as ROM (Read Only Memory) or RAM (Random Access Memory), and stores the program and data used for processing. Memory 1802 may also operate as the storage unit 318 in Figure 3.

[0142] The CPU 1801 (processor) operates as the selection unit 111, estimation unit 112, and search unit 113 in Figure 1, for example, by executing a program using the memory 1802. The CPU 1801 also operates as the analysis unit 311, training unit 312, generation unit 313, selection unit 314, estimation unit 315, and search unit 316 in Figure 3, by executing a program using the memory 1802.

[0143] The input device 1803 is, for example, a keyboard, a pointing device, etc., and is used for inputting instructions or information from the user or operator. The output device 1804 is, for example, a display device, a printer, etc., and is used for inquiries or instructions to the user or operator, and for outputting processing results. The processing results may be search results 333. The output device 1804 may operate as the output unit 114 in Figure 1 or the display unit 317 in Figure 3.

[0144] The auxiliary storage device 1805 is, for example, a magnetic disk drive, an optical disk drive, a magneto-optical disk drive, a tape drive, etc. The auxiliary storage device 1805 may also be a hard disk drive. The information processing device can store programs and data in the auxiliary storage device 1805 and load them into the memory 1802 for use. The auxiliary storage device 1805 may also operate as the storage unit 318 in Figure 3.

[0145] The media drive unit 1806 drives the portable recording medium 1809 and accesses its recorded contents. The portable recording medium 1809 is a memory device, flexible disk, optical disk, magneto-optical disk, etc. The portable recording medium 1809 may also be a CD-ROM (Compact Disk Read Only Memory), DVD (Digital Versatile Disk), USB (Universal Serial Bus) memory, etc. The user or operator can store programs and data on the portable recording medium 1809 and load them into the memory 1802 for use.

[0146] Thus, the computer-readable recording medium that stores the programs and data used in the processing is a physical (non-temporary) recording medium such as memory 1802, auxiliary storage device 1805, or portable recording medium 1809.

[0147] The network connection device 1807 is a communication interface circuit that connects to a communication network such as a WAN (Wide Area Network) or LAN (Local Area Network) and performs data conversion associated with communication. The information processing device can receive programs and data from external devices via the network connection device 1807 and load them into the memory 1802 for use. The network connection device 1807 may also operate as the output unit 114 in Figure 1.

[0148] Note that the information processing device does not need to include all the components shown in Figure 18, and some components may be omitted depending on the intended use or conditions of the information processing device. For example, if the portable recording medium 1809 or a communication network is not used, the media drive device 1806 or the network connection device 1807 may be omitted.

[0149] While embodiments of the disclosure and their advantages have been described in detail, those skilled in the art will be able to make various modifications, additions, and omissions without departing from the scope of the invention as expressly stated in the claims.

[0150] With reference to Figures 1 to 18, the following additional information is disclosed regarding the embodiments described. (Note 1) Based on the first characteristic information that shows the characteristics of the target of intervention, a specific intervention method is selected from among multiple intervention methods. An estimated change in characteristics is calculated, representing the difference between the second characteristic information, which shows the characteristics of the subject after the specific intervention method has been applied to the subject of the intervention, and the first characteristic information. From case data including intervention methods, characteristic information, and characteristic change amounts for each of multiple individuals, a specific case data is searched for that includes a combination of characteristic information and characteristic change amounts similar to the combination of the first characteristic information and the estimated characteristic change amount, and the specific intervention method. Outputs information based on the aforementioned specific case data. A search program that causes a computer to perform a process. (Note 2) The search program according to Appendix 1, characterized in that the computer further performs a process to generate the first characteristic information based on information obtained from the intervention subject for each of the multiple items. (Note 3) The information obtained from the intervention subjects for each of the aforementioned multiple items includes the numerical values ​​for each of the aforementioned multiple items, The first characteristic information includes numerical values ​​for each of the multiple factors related to the multiple items, The search program described in Appendix 2, characterized in that the numerical value of each of the aforementioned multiple factors is a weighted sum of the numerical values ​​of each of the aforementioned multiple items. (Note 4) The case data, which includes the intervention method, characteristic information, and characteristic change amount for each of the aforementioned multiple individuals, further includes situational information indicating the circumstances when the intervention was carried out using the intervention method for each of the aforementioned multiple individuals. The search program according to Appendix 1, characterized in that the information based on the specific case data includes situational information contained in the specific case data. (Note 5) The process of selecting the aforementioned specific intervention method is: A process to obtain first state information indicating the state of the person to be intervened, based on the first characteristic information, A process to obtain second state information indicating the state of the person after each of the multiple intervention methods has been applied to the person, based on the first characteristic information and each of the multiple intervention methods; A process of selecting a specific intervention method from among the multiple intervention methods based on the first state information and the second state information, A search program according to any one of the appendices 1 to 4, characterized by including the following: (Note 6) The process for obtaining the first state information includes a process for obtaining the first state information from the first characteristic information using a state estimation model. The process for obtaining the second state information is as follows: A process using a characteristic estimation model to obtain third characteristic information that shows the characteristics of the target person after each of the multiple intervention methods has been applied to them, from the first characteristic information and each of the multiple intervention methods, A process to obtain the second state information from the third characteristic information using the state estimation model, Includes, The state estimation model is generated by machine learning using training data, which includes multiple characteristic information and state information corresponding to each of the multiple characteristic information. The search program described in Appendix 5 is characterized in that the characteristic estimation model is generated by machine learning using training data and the state estimation model, the characteristic estimation model comprising a plurality of characteristic pieces of information, an intervention method corresponding to each of the plurality of characteristic pieces of information, and state information corresponding to a combination of each of the plurality of characteristic pieces of information and the intervention method corresponding to each of the plurality of characteristic pieces of information. (Note 7) The process for determining the estimated characteristic change is as follows: Using the characteristic estimation model, a process is performed to obtain second characteristic information from the first characteristic information and the specific intervention method, A process for determining the estimated characteristic change amount using the first characteristic information and the second characteristic information, A search program as described in Appendix 6, characterized by including the following: (Note 8) A selection unit that selects a specific intervention method from among multiple intervention methods based on first characteristic information that shows the characteristics of the person to be intervened with, An estimation unit that calculates an estimated change in characteristics, which represents the difference between second characteristic information showing the characteristics after the specific intervention method has been applied to the subject of the intervention, and first characteristic information. A search unit searches for specific case data from case data including intervention methods, characteristic information, and characteristic change amounts for each of multiple individuals, including a combination of characteristic information and characteristic change amount similar to the combination of the first characteristic information and the estimated characteristic change amount, and the specific intervention method. An output unit that outputs information based on the aforementioned specific case data, A search device characterized by comprising the following features. (Note 9) The search device according to Appendix 8, further comprising a generation unit that generates the first characteristic information based on information obtained from the intervention subject for each of the multiple items. (Note 10) The information obtained from the intervention subjects for each of the aforementioned multiple items includes the numerical values ​​for each of the aforementioned multiple items, The first characteristic information includes numerical values ​​for each of the multiple factors related to the multiple items, The search device according to Appendix 9, characterized in that the numerical value of each of the aforementioned multiple factors is a weighted sum of the numerical values ​​of each of the aforementioned multiple items. (Note 11) The case data, which includes the intervention method, characteristic information, and characteristic change amount for each of the aforementioned multiple individuals, further includes situational information indicating the circumstances when the intervention was carried out using the intervention method for each of the aforementioned multiple individuals. The search device according to Appendix 8, characterized in that the information based on the specific case data includes situational information contained in the specific case data. (Note 12) The search device according to any one of the appendices 8 to 11, characterized in that the selection unit obtains first state information indicating the state of the person to be intervened corresponding to the first characteristic information based on the first characteristic information, obtains second state information indicating the state of the person to be intervened after each of the multiple intervention methods has been applied to the person to be intervened based on the first characteristic information and each of the multiple intervention methods, and selects the specific intervention method from the multiple intervention methods based on the first state information and the second state information. (Note 13) The selection unit obtains first state information from the first characteristic information using a state estimation model, obtains third characteristic information from the first characteristic information and each of the multiple intervention methods using the characteristic estimation model, which indicates the characteristics after each of the multiple intervention methods has been applied to the person to be intervened, obtains second state information from the third characteristic information using the state estimation model, The state estimation model is generated by machine learning using training data, which includes multiple characteristic information and state information corresponding to each of the multiple characteristic information. The search device according to Appendix 12, characterized in that the characteristic estimation model is generated by machine learning using training data and the state estimation model, the characteristic estimation model comprising the plurality of characteristic information, an intervention method corresponding to each of the plurality of characteristic information, and state information corresponding to a combination of each of the plurality of characteristic information and the intervention method corresponding to each of the plurality of characteristic information. (Note 14) Based on the first characteristic information that shows the characteristics of the target of intervention, a specific intervention method is selected from among multiple intervention methods. An estimated change in characteristics is calculated, representing the difference between the second characteristic information, which shows the characteristics of the subject after the specific intervention method has been applied to the subject of the intervention, and the first characteristic information. From case data including intervention methods, characteristic information, and characteristic change amounts for each of multiple individuals, a specific case data is searched for that includes a combination of characteristic information and characteristic change amounts similar to the combination of the first characteristic information and the estimated characteristic change amount, and the specific intervention method. Outputs information based on the aforementioned specific case data. A search method characterized by having a computer perform the processing. (Note 15) The search method according to Appendix 14, characterized in that the computer further performs a process to generate the first characteristic information based on information obtained from the intervention subject for each of the multiple items. (Note 16) The information obtained from the intervention subjects for each of the aforementioned multiple items includes the numerical values ​​for each of the aforementioned multiple items, The first characteristic information includes numerical values ​​for each of the multiple factors related to the multiple items, The search method according to Appendix 15, characterized in that the numerical value of each of the aforementioned multiple factors is a weighted sum of the numerical values ​​of each of the aforementioned multiple items. (Note 17) The case data, which includes the intervention method, characteristic information, and characteristic change amount for each of the aforementioned multiple individuals, further includes situational information indicating the circumstances when the intervention was carried out using the intervention method for each of the aforementioned multiple individuals. The search method according to Appendix 14, characterized in that the information based on the specific case data includes situational information contained in the specific case data. (Note 18) The process of selecting the aforementioned specific intervention method is: A process to obtain first state information indicating the state of the person to be intervened, based on the first characteristic information, A process to obtain second state information indicating the state of the person after each of the multiple intervention methods has been applied to the person, based on the first characteristic information and each of the multiple intervention methods; A process of selecting a specific intervention method from among the multiple intervention methods based on the first state information and the second state information, A search method according to any one of the appendices 14 to 17, characterized by including the following. (Note 19) The process for obtaining the first state information includes a process for obtaining the first state information from the first characteristic information using a state estimation model. The process for obtaining the second state information is as follows: A process using a characteristic estimation model to obtain third characteristic information that shows the characteristics of the target person after each of the multiple intervention methods has been applied to them, from the first characteristic information and each of the multiple intervention methods, A process to obtain the second state information from the third characteristic information using the state estimation model, Includes, The state estimation model is generated by machine learning using training data, which includes multiple characteristic information and state information corresponding to each of the multiple characteristic information. The search method according to Appendix 18, characterized in that the characteristic estimation model is generated by machine learning using training data and the state estimation model, the characteristic estimation model comprising the plurality of characteristic information, an intervention method corresponding to each of the plurality of characteristic information, and state information corresponding to a combination of each of the plurality of characteristic information and the intervention method corresponding to each of the plurality of characteristic information. [Explanation of Symbols]

[0151] 101, 301 Search device 111, 314 Selection section 112, 315 Estimation part 113, 316 Search section Output section of 114 311 Analysis Department 312 Training Department 313 Generation part 317 Display section 318 Storage section 321, 328 Survey data 322 Coefficient Information 323, 324 Training data 325, 601 State Estimation Models 326, 901 Factor Score Estimation Models 327 case collection Factor score vectors 329, 331 330 Information on outreach 332 Estimated Difference Vectors 333 search results 611, 911~913 Input Layer 612, 914 output layers Nodes 621-1~621-5, 622-1~622-5, 921-1~921-5, 922-1~922-K, 923-1~923-5, 924-1~924-5 1001 Combination Model 1011 DNN 1801 CPU 1802 memory 1803 Input device 1804 Output device 1805 Auxiliary storage 1806 Media drive device 1807 Network Connection Device 1808 Bus 1809 Portable recording media

Claims

1. Based on the first characteristic information that shows the characteristics of the person to be intervened with, a specific intervention method is selected from among several intervention methods. Second characteristic information is estimated that shows the characteristics when the specific intervention method is applied to the aforementioned target person. An estimated characteristic change amount is calculated, which represents the difference between the second characteristic information and the first characteristic information. From case data including intervention methods for each of multiple target individuals, fourth characteristic information indicating the characteristics of each of the multiple target individuals, and characteristic change amounts indicating the amount of change in the fourth characteristic information, a specific case data is searched for that includes a combination of characteristic information and characteristic change amounts similar to the combination of the first characteristic information and the estimated characteristic change amount, and the specific intervention method. Outputs information based on the aforementioned specific case data. A search program that causes a computer to perform a process.

2. The search program according to claim 1, characterized in that the computer is further instructed to perform a process of generating the first characteristic information based on information obtained from the intervention subject for each of the multiple items.

3. The information obtained from the intervention subjects for each of the aforementioned multiple items includes the numerical values ​​for each of the aforementioned multiple items, The first characteristic information includes the numerical values ​​of each of the multiple factors related to the multiple items, The search program according to claim 2, characterized in that the numerical value of each of the aforementioned multiple factors is a weighted sum of the numerical values ​​of each of the aforementioned multiple items.

4. The case data further includes situational information indicating the circumstances when an intervention was carried out using the intervention method for each of the plurality of target persons, The search program according to claim 1, characterized in that the information based on the specific case data includes situational information contained in the specific case data.

5. The process of selecting the aforementioned specific intervention method is: A process to obtain first state information indicating the state of the person to be intervened, based on the first characteristic information, A process to obtain second state information indicating the state of the person after each of the multiple intervention methods has been applied to the person subject to intervention, based on the first characteristic information, A process of selecting a specific intervention method from among the multiple intervention methods based on the first state information and the second state information, A search program according to any one of claims 1 to 4, characterized by including the following:

6. The process for obtaining the first state information includes a process for obtaining the first state information from the first characteristic information using a state estimation model. The process for obtaining the second state information is as follows: A process using a characteristic estimation model to obtain third characteristic information from the first characteristic information, which shows the characteristics after each of the multiple intervention methods has been applied to the target person for intervention, A process to obtain the second state information from the third characteristic information using the state estimation model, Includes, The state estimation model is generated by machine learning using training data, which includes multiple characteristic information and state information corresponding to each of the multiple characteristic information. The characteristic estimation model includes the plurality of characteristic information and the corresponding information for each of the plurality of characteristic information. The search program according to claim 5, characterized in that it is generated by machine learning using training data, which includes an input method, state information corresponding to a combination of each of the plurality of characteristic information and an intervention method corresponding to each of the plurality of characteristic information, and the state estimation model.

7. The process for determining the estimated characteristic change is as follows: Using the aforementioned characteristic estimation model, a process is performed to obtain the second characteristic information from the first characteristic information, A process for determining the estimated characteristic change amount using the first characteristic information and the second characteristic information, A search program according to claim 6, characterized by including the following:

8. A selection unit that selects a specific intervention method from among multiple intervention methods based on first characteristic information that shows the characteristics of the person to be intervened with, An estimation unit estimates second characteristic information that shows the characteristics when the specific intervention method is applied to the subject of the intervention, and calculates an estimated characteristic change amount that represents the difference between the second characteristic information and the first characteristic information. A search unit searches for specific case data from case data including intervention methods for each of multiple target individuals, fourth characteristic information indicating the characteristics of each of the multiple target individuals, and characteristic change amounts indicating the amount of change in the fourth characteristic information, to find combinations of characteristic information and characteristic change amounts similar to the combination of the first characteristic information and the estimated characteristic change amount, and the specific intervention method. An output unit that outputs information based on the aforementioned specific case data, A search device characterized by comprising the following features.

9. Based on the first characteristic information that shows the characteristics of the person to be intervened with, a specific intervention method is selected from among several intervention methods. Second characteristic information is estimated that shows the characteristics when the specific intervention method is applied to the aforementioned target person. An estimated characteristic change amount is calculated, which represents the difference between the second characteristic information and the first characteristic information. From case data including intervention methods for each of multiple target individuals, fourth characteristic information indicating the characteristics of each of the multiple target individuals, and characteristic change amounts indicating the amount of change in the fourth characteristic information, a specific case data is searched for that includes a combination of characteristic information and characteristic change amounts similar to the combination of the first characteristic information and the estimated characteristic change amount, and the specific intervention method. Outputs information based on the aforementioned specific case data. A search method characterized by having a computer perform the processing.

Images