JP7245125B2 - Generation device, generation method, and generation program - Google Patents

Description

The present invention relates to a generation device, generation method, and generation program for generating data.

In life insurance enrollment screening, future onset and hospitalization risks are predicted based on the health condition of the applicant. Health conditions are represented by multivariate data such as health checkup results and notification information. Furthermore, when considering changes in health conditions, the number of dimensions of multivariate data becomes even more enormous because risk prediction is performed in consideration of health conditions for multiple years.

Techniques for time-series data analysis include vector autoregression models and LSTM (Long short-term memory), and there are methods of analysis using regression models and neural networks as independent variables at each time point.

Further, Patent Literature 1 discloses a data analysis device that analyzes health conditions for multiple years. This data analysis device stores quantitative data and qualitative data each having an ID and time information, generates time-series quantitative event data from the quantitative data, generates time-series qualitative event data from the qualitative data, and generates time-series qualitative event data from the qualitative data. Extracting a characteristic portion having a change from one of series quantitative and qualitative event data, generating a time-series event pattern from a set of event data corresponding to the characteristic portion, generating an ID included in the time-series event pattern and the time An ID included in the other of the series quantitative and qualitative event data is associated, and the associated time series event pattern and the other of the associated time series quantitative and qualitative event data are displayed.

JP-A-2006-285672

However, when extracting feature amounts necessary for analysis by exhaustive search, feature amounts unnecessary for analysis are also generated, increasing calculation costs. In addition, there is a possibility that feature amounts unnecessary for analysis may adversely affect the intended analysis.

An object of the present invention is to realize efficient and highly accurate data analysis.

A generation device that is one aspect of the invention disclosed in the present application is a generation device that includes a processor that executes a program and a storage device that stores the program, and exhibits chronological characteristics obtained from chronological data. It is possible to access temporal characteristic information and grouping information that defines a plurality of groups to which the temporal data should belong, and the processor, based on the temporal characteristic information, selects, from the temporal data to be analyzed, the a generation process of generating a plurality of pieces of temporal feature data indicating characteristics over time; a division process of dividing the plurality of pieces of temporal feature data generated by the generation processing into the plurality of groups based on the grouping information; and dimension compression processing for dimensionally compressing each of the plurality of groups divided by the division processing, and dimensionally compressing the non-temporal data to be analyzed .

According to representative embodiments of the present invention, efficient and highly accurate data analysis can be achieved. Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

FIG. 1 is a block diagram showing a hardware configuration example of a generation device. FIG. 2 is a block diagram of a functional configuration example of the generation device according to the first embodiment; FIG. 3 is an explanatory diagram showing an example of the contents of notification information stored. FIG. 4 is an explanatory diagram showing an example of the contents of domain knowledge stored. FIG. 5 is an explanatory diagram showing an example of temporal features. FIG. 6 is an explanatory diagram showing an example of divided temporal features. FIG. 7 is an explanatory diagram showing an example of high-order feature data. FIG. 8 is an explanatory diagram showing an input/output screen example 1. FIG. FIG. 9 is an explanatory diagram showing an example of a notification information input screen. FIG. 10 is an explanatory diagram showing example 2 of the input/output screen. FIG. 11 is a flowchart illustrating an example of a generation processing procedure by the generation device. FIG. 12 is a block diagram of a functional configuration example of a generation device according to a second embodiment; FIG. 13 is an explanatory diagram showing an example of a multimodal neural network. FIG. 14 is an explanatory diagram showing an example of an input/output screen showing analysis results by the multimodal neural network.

Hereinafter, a generation device according to the present invention will be described with reference to the accompanying drawings. Provided herein are examples of claim risk prediction in life insurance underwriting. In the underwriting assessment, based on the information notified by the applicant (hereinafter referred to as notification information), the future insurance claim payment risk is assessed and the approval or refusal of insurance enrollment is determined. The notification information includes test results of physical examinations, medical interviews, medical history, and the like.

<Hardware configuration example of generation device>
FIG. 1 is a block diagram showing a hardware configuration example of a generation device. The generation device 100 has a processor 101 , a storage device 102 , an input device 103 , an output device 104 and a communication interface (communication IF) 105 . Processor 101 , storage device 102 , input device 103 , output device 104 and communication IF 105 are connected by bus 106 . A processor 101 controls the generator 100 . Storage device 102 serves as a work area for processor 101 . Also, the storage device 102 is a non-temporary or temporary recording medium that stores various programs and data. Examples of the storage device 102 include ROM (Read Only Memory), RAM (Random Access Memory), HDD (Hard Disk Drive), and flash memory. The input device 103 inputs data. The input device 103 includes, for example, a keyboard, mouse, touch panel, numeric keypad, and scanner. The output device 104 outputs data. Output devices 104 include, for example, displays and printers. Communication IF 105 connects to a network and transmits and receives data.

<Functional Configuration Example of Generation Device 100>
FIG. 2 is a block diagram of a functional configuration example of the generation device 100 according to the first embodiment. The generation device 100 includes a determination unit 201 , a generation unit 202 , a division unit 203 , a dimension compression unit 204 , a combination unit 205 and an analysis unit 206 . Specifically, the determination unit 201, the generation unit 202, the division unit 203, the dimension compression unit 204, the combination unit 205, and the analysis unit 206, for example, convert the program stored in the storage device 102 shown in FIG. It is realized by executing

Moreover, the generation device 100 only needs to have at least the generation unit 202, the division unit 203, and the dimension compression unit 204, and the determination unit 201, the combination unit 205, and the analysis unit 206 can communicate with the generation device 100. It may also be implemented on other computers.

The generating device 100 also stores the notification information 300 and the domain knowledge 400 in the storage device 102 . The announcement information 300 and the domain knowledge 400 may be stored in the generation device 100 in advance, or may be obtained from another computer that can communicate with the generation device 100 . First, the notification information 300 will be described in detail.

FIG. 3 is an explanatory diagram showing an example of the contents of the notification information 300. As shown in FIG. The notification information 300 is information necessary for the insurance contract notified by the contract applicant, and is data to be analyzed. Notification information 300 includes notification basic information 310 , health checkup results 320 , interview results 330 , and past history 340 . The basic notification information 310 is basic information regarding the notification of the contract applicant. Notification basic information 310 includes name ID 311 , date of birth 312 , and age 313 .

The name ID 311 is identification information that uniquely identifies the contract applicant. The date of birth 312 is the date on which the contract applicant was born. The three entries of the contract applicant whose name ID 311 is "0001" in FIG. 3 indicate the analysis target data for the past three years of the contract applicant. The age 313 is the elapsed years in years calculated from the date of birth 312 of the contract applicant. In the example described later, for three entries of a contract applicant whose name ID 311 is "0001", age 313 is "47" for the first year in chronological order, "48" for the second year in chronological order, and "49" for age 313. This is the third year in chronological order.

The health checkup result 320 is the result of the health checkup that the contract candidate received. The health checkup results 320 include body weight 321 , BMI (Body Mass Index) 322 , systolic blood pressure 323 , diastolic blood pressure 324 , and fasting blood sugar 325 . The weight 321 is the weight of the contract applicant's body. BMI322 is a body mass index representing the degree of obesity of a person and is calculated by weight/(height ² ). BMI 322 indicates that the smaller the value, the thinner the person, and the larger the value, the fatter the person.

Systolic blood pressure 323 is the pressure exerted on the wall of the aorta by the blood pushed out by contraction of the heart in the state of pumping blood from the heart to the aorta. The diastolic blood pressure 324 is the pressure applied to the aortic vessel wall, which is lowered when blood flows from the aorta into the heart due to the dilation of the heart and the amount of blood in the aorta decreases when the blood returns to the heart. Fasting blood glucose 325 is the blood glucose level measured in the fasting state.

The inquiry result 330 is the result of the inquiry received by the contract applicant. The interview results 330 include smoking habits 331 , drinking habits 332 , and exercise habits 333 . The smoking habit 331 is the presence/absence, frequency, and amount of smoking of the person who wishes to make a contract. The drinking habit 332 is the presence or absence of drinking, the frequency of drinking, and the amount of drinking of the contract applicant. The exercise habit 333 is the presence/absence, frequency, and amount of exercise of the contract applicant.

The anamnesis 340 is a history of the contract applicant having already received or been hospitalized. The medical history 340 includes hypertension medical history 341 , hypertension hospitalization history 342 , and diabetes medical history 343 . The hypertension medical examination history 341 is a history of medical examinations regarding hypertension by the contract applicant. The hypertension hospitalization history 342 is a history of hospitalization for hypertension of the contract applicant. Diabetes consultation history 343 is a history of consultations regarding diabetes by the contract applicant.

FIG. 4 is an explanatory diagram showing an example of the contents of the domain knowledge 400. As shown in FIG. The domain knowledge 400 is qualitative information for various types of information included in the notification information 300, such as the health checkup result 320, medical interview result 330, and medical history 340, and corresponds to medical knowledge. Specifically, for example, domain knowledge 400 includes temporal data determination knowledge 410 , temporal feature knowledge 420 , and temporal feature segmentation knowledge 430 .

The chronological data judgment knowledge 410 is judgment information for judging to which of the non-chronological data 231 and the chronological data 232 the notification information 300 of the contract applicant, which is the data to be analyzed, corresponds. Specifically, for example, the chronological data determination knowledge 410 defines items of the notification information 300 corresponding to the non-chronological data 231 and the chronological data 232 .

Non-chronological data items 411 include items corresponding to non-chronological data 231 . The non-temporal data 231 is data that does not change in time series, or data that changes but is meaningless. Non-chronological data 231 includes, for example, smoking habits 331, drinking habits 332, various medical examination histories, and various hospitalization histories. The chronological data items 412 include items corresponding to the chronological data 232 . The chronological data 232 is time-series data whose chronological change is significant. The chronological data 232 includes, for example, weight 321, BMI 322, systolic blood pressure 323, diastolic blood pressure 324, fasting blood sugar 325, and the like.

The temporal feature knowledge 420 is temporal feature information indicating temporal features (temporal features) obtained from the temporal data 232 . For example, the feature knowledge over time 420 defines a basic statistic item 421, a change amount item 422, a change rate item 423, and so on. The basic statistics items 421 include items corresponding to basic statistics obtained from the temporal data 232 . Basic statistics are, for example, the maximum value, minimum value, and average value of the temporal data 232 .

The amount of change item 422 is an item corresponding to the amount of change that indicates the change in two consecutive values in the temporal data 232 . For example, if the temporal data 232 is yearly weight 321 for one to three years, the amount of change (1st, 2nd year) is the difference between the weight 321 in the first year and the weight 321 in the second year. , the amount of change (second and third years) is the difference between the weight 321 in the second year and the weight 321 in the third year.

The change rate item 423 is an item corresponding to a value indicating the rate of change between two consecutive values in the temporal data 232 . For example, if the temporal data 232 is yearly weight 321 for one to three years, the rate of change (1st, 2nd year) is the difference between the weight 321 in the first year and the weight 321 in the second year. It is the value divided by the weight 321 in the year, and the amount of change (2nd and 3rd years) is the difference between the weight 321 in the 2nd year and the weight 321 in the 3rd year divided by the weight 321 in the 2nd year. is.

The temporal feature division knowledge 430 is grouping information defining groups to which the temporal data 232 should belong. The temporal feature segmentation knowledge 430 is specifically defined based on, for example, statistical knowledge or medical knowledge. Specifically, for example, the temporal feature division knowledge includes a body type basic information item 431, a blood pressure system test value item 432, a blood glucose system test value item 433, a liver function system test value item 434, and so on. Including as

The body type basic information item 431 includes items corresponding to body type basic information. The body shape basic information is basic information about the body shape of the contract applicant. The body type basic information item 431 includes, for example, age 313, weight 321, and BMI 322 as items.

The blood pressure test value item 432 includes items corresponding to blood pressure test values. Blood pressure test values are test values related to the blood pressure of the contract applicant. The blood pressure test value item 432 includes, for example, the systolic blood pressure 323 and the diastolic blood pressure 324 as items.

The blood sugar test value item 433 includes items corresponding to blood sugar test values. The blood sugar test value is a blood sugar test value of the contract applicant. The blood sugar test value item 433 includes, for example, fasting blood sugar 325 and HbA1c as items.

The liver function test value item 434 includes items corresponding to liver function test values. The liver function system test value is a test value related to the liver function of the contract applicant. The liver function test value item 434 includes, for example, GOT (glutamate oxaloacetate transaminase), GPT (glutamate pyruvate transaminase), and γ-GTP (γ glutamyltranspeptidase) as items.

Returning to FIG. 2 , the determination unit 201 determines whether the notification information 300 corresponds to the temporal data 232 or the non-temporal data 231 based on the temporal data determination knowledge 410 . An entry in which the name ID 311 in the notification information 300 in FIG. 3 is "0001" is taken as an example. Age 313 , weight 321 , BMI 322 , systolic blood pressure 323 , diastolic blood pressure 324 and fasting blood glucose 325 are included in chronological data items 412 . Therefore, the determination unit 201 determines that the age 313 is "47", "48" and "49", the weight 321 is "83.4", "86.6" and "92.0", and the BMI 322 is "22.8". , "24.3", "26.0", "124.9", "128.5", "133.8" of the systolic blood pressure 323, "80.7", "86. 1”, “90.0”, and “104.5”, “107.2”, and “110.0” of the fasting blood sugar 325 are the chronological data 232 of the contract applicant whose name ID 311 is “0001”. , and determine. The determination unit 201 outputs determined temporal data 232 as a determination result.

Smoking habits 331 , drinking habits 332 , exercise habits 333 , hypertension medical history 341 , hypertension hospitalization history 342 , and diabetes medical history 343 are included in non-chronological data items 411 . Therefore, the determining unit 201 selects “none”, “none”, and “none” for the smoking habit 331, “1st week”, “1st week”, and “2nd week” for the drinking habit 332, and “1st week” for the exercise habit 333. , "week 1", "none", hypertension medical history 341 "none", "none", "none", hypertension hospitalization history 342 "none", "none", "none", diabetes medical history 343 are the non-temporal data 231 of the contract applicant whose name ID 311 is "0001". The determination unit 201 outputs the determined non-temporal data 231 as a determination result.

The generation unit 202 generates temporal feature data representing temporal features from the temporal data 232 based on the temporal feature knowledge 420 . The generation unit 202 calculates the maximum value 511, the minimum value 512, the average value, . The generation unit 202 calculates the amount of change (1st and 2nd years) 521, the amount of change (2nd and 3rd years) 522, . The generation unit 202 calculates a change rate (1st and 2nd years) 531, a change rate (2nd and 3rd years) 532, .

An entry in which the name ID 311 in the notification information 300 in FIG. 3 is "0001" is taken as an example. For example, in the case of "83.4", "86.6", and "92.0" for the weight 321 included in the chronological data 232, the maximum value 511 is "92.0", which is the weight 321 in the third year. The minimum value 512 is ``83.4'', which is the weight 321 in the first year, and the average value is ``87.3'', which is the average of ``83.4'', ``86.6'', and ``92.0''. be.

Further, the amount of change (1st and 2nd years) 521 is "3.2" obtained by subtracting "83.4" from "86.6", and the amount of change (2nd and 3rd years) 522 is "92.0". ' minus '86.6' is '5.4'. Also, the rate of change (first and second years) 531 is obtained by dividing the change amount (first and second years) 521 of "3.2" by the weight 321 of the first year of "83.4", which is "0. 04", the rate of change (2nd and 3rd years) 532 is "0 .06". The same applies to BMI 322, systolic blood pressure 323, diastolic blood pressure 324, fasting blood sugar 325, etc. other than body weight 321. FIG. The generation unit 202 outputs the generation result as temporal feature data 500 .

FIG. 5 is an explanatory diagram showing an example of temporal feature data 500. As shown in FIG. The chronological feature data 500 includes a basic statistic 510 , a change amount 520 and a change rate 530 . The basic statistics 510 include a maximum value 511, a minimum value 512, an average value (not shown), . . . calculated according to the basic statistics item 421. FIG. The amount of change 520 includes an amount of change (1st and 2nd years) 521, an amount of change (2nd and 3rd years) 522, . . . The rate of change 530 includes a rate of change (1st and 2nd years) 531, a rate of change (2nd and 3rd years) 532, . . .

The basic statistic 510, the amount of change 520, and the rate of change 530 are each generated by the generator 202 for each type of temporal data 232. FIG. Entries 501-1, 501-2, . . . , 502-1, 502-2, . , is chronological feature data relating to a certain contract applicant (for example, a contract applicant whose name ID 311 is "0001"). , 502-1, 502-2, . . . , 501-3, 503-2, . to generate

Entry 501 is characteristic data over time according to body type basic information item 431 . Specifically, for example, entry 501 - 1 is feature data over time indicating basic statistics, amount of change, and rate of change for body weight 321 . Entry 501-2 is feature data over time indicating basic statistics, amount of change, and rate of change of BMI 322. FIG.

Entry 502 is characteristic data over time according to the blood pressure system test value item 432 . Specifically, entry 502 - 1 , for example, is feature data over time indicating basic statistics, amount of change, and rate of change for systolic blood pressure 323 . Entry 502 - 2 is feature data over time indicating basic statistics, amount of change, and rate of change for diastolic blood pressure 324 .

Entry 503 is chronological feature data according to blood glucose test value item 433 . Specifically, entry 503 - 1 , for example, is feature data over time indicating basic statistics, amount of change, and rate of change for fasting blood glucose 325 . Entry 503-2 is chronological feature data indicating basic statistics, amount of change, and rate of change for HbA1c.

Entry 504 is characteristic data over time according to liver function test value item 434 . Specifically, for example, entry 504-1 is chronological feature data indicating basic statistics, amount of change, and rate of change for GOT. Entry 504-2 is feature data over time indicating basic statistics, amount of change, and rate of change for GPT.

Returning to FIG. 2 , the dividing unit 203 divides the plurality of temporal feature data generated by the generating unit 202 into a plurality of groups based on the temporal feature division knowledge 430 and outputs divided temporal feature data 600 . , 502-1, 502-2, . . . , 501-3, 503-2, . , 504-2, . . . The multiple groups are divided temporal feature data 600-1, 600-2, . . . , 600-n (n is an integer of 1 or more). Each of the divided temporal feature data 600-1, 600-2, . 433, liver function test value items 434, . . .

The temporal feature segmentation knowledge 430 is defined based on statistical knowledge or medical knowledge. (Example 1) BMI 322 is an index calculated from height and weight 321 . There is not much change in height among the age groups of those who wish to purchase insurance. Therefore, changes in BMI 322 have a very strong correlation with changes in body weight 321 . If there are multiple highly correlated items in the notification information 300, they become redundant information and cause inefficiency in data analysis.

On the other hand, by applying the temporal feature division knowledge 430, the temporal feature data of a plurality of items including redundancy (age 313, weight 321, BMI 322, . -1 group (hereinafter referred to as body type group 600-1). As a result, the generation device 100 can collect the temporal feature data generated using the values of the plurality of items as a body shape group, perform dimension compression, and extract high-dimensional features (hereinafter referred to as high-order features). This makes it possible to improve the efficiency of data analysis.

(Example 2) Both the systolic blood pressure 323 and the diastolic blood pressure 324 are known to slowly deteriorate with aging. In addition, even when the blood pressure decreases or increases due to improvement or deterioration of lifestyle habits, the systolic blood pressure 323 and the diastolic blood pressure 324 decrease or increase while maintaining a balance. However, when this balance changes, it is said to be a sign of hypertensive diseases such as arteriosclerosis.

Therefore, by applying the temporal feature division knowledge 430, the temporal feature data of the systolic blood pressure 323 and the diastolic blood pressure 324 are divided into a group of the divided temporal feature data 600-2 called the blood pressure test value item 432 (hereinafter referred to as the blood pressure group). 600-2). As a result, the generating apparatus 100 collects the temporal feature data generated using the values of the systolic blood pressure 323 and the diastolic blood pressure 324 as a blood pressure group, dimensionally compresses them, and extracts higher-order features. , it is possible to obtain the composite feature quantity necessary for data analysis.

For the same reason, the fasting blood sugar 325 and HbA1c chronological feature data are grouped into a group of divided chronological feature data 600-3 (hereafter, blood glucose group 600-3) called blood glucose test value item 433, and GOT, GPT , γ-GTP are grouped into a group of divided temporal feature data 600-4 called liver function test value items 434 (hereafter, liver function group 600-4).

FIG. 6 is an explanatory diagram showing an example of divided temporal feature data 600. As shown in FIG. The segmented temporal features include a body type group 600-1, a blood pressure group 600-2, a blood sugar group 600-3, and a liver function group 600-4. The figure group 600 - 1 is a data set containing chronological characteristic data corresponding to the figure basic information item 431 . A blood pressure group 600 - 2 is a data set containing chronological characteristic data corresponding to the blood pressure test value item 432 . The glycemic group 600-3 is a data set containing chronological feature data corresponding to the blood glucose test value item 433. FIG. The hepatic function group 600-4 is a data set containing chronological characteristic data corresponding to the hepatic function test value item 434. FIG.

Body type group 600-1 includes weight 321 feature data over time 601-1 and BMI 322 over time feature data 601-2. A data string of the temporal feature data 601-1 of the body weight 321 is assumed to be a vector Ub1-1. A data string of the temporal characteristic data 601-2 of the BMI 322 is assumed to be a vector Ub1-2.

The blood pressure system group 600-2 includes temporal feature data 602-1 of the systolic blood pressure 323 and temporal feature data 602-2 of the diastolic blood pressure 324. FIG. Let the data string of the temporal feature data 602-1 of the systolic blood pressure 323 be a vector Ub2-1. Let the data string of the temporal feature data 602-2 of the diastolic blood pressure 324 be a vector Ub2-2.

The blood glucose system group 600-3 includes fasting blood glucose 325 time course feature data 603-1 and HbA1c time course feature data 602-2. Let the data string of the fasting blood glucose 325 chronological feature data 603-1 be a vector Ub3-1. Let the data string of the HbA1c chronological feature data 602-2 be a vector Ub3-2.

The liver function system group 600-4 includes GOT feature data over time 604-1 and GPT feature data over time 604-2. A data string of the GOT temporal feature data 604-1 is assumed to be a vector Ub4-1. A data string of the GPT chronological feature data 604-2 is assumed to be a vector Ub4-2.

Returning to FIG. 2 , the dimension compression unit 204 dimensionally compresses the input data to generate high-order feature data 700 . Specifically, for example, the dimension compression unit 204 dimension-compresses each of the plurality of groups divided by the division unit 203 . Specifically, for example, the dimension compression unit 204 performs dimension compression on each of the divided temporal feature data 600-1 to 600-n to generate high-order feature data 702-1 to 702-n regarding the temporal data 232. .

The dimension compression unit 204 also performs dimension compression on the non-temporal data 231 to generate high-order feature data 701 on the non-temporal data 231 . Extraction of high-order feature data by dimensional compression of data is realized by known dimensional compression methods such as principal components analysis (PCA) and stacked autoencoder. Further, the dimension compression unit 204 may generate high-order feature data by once expanding the number of dimensions of data using a neural network or the like, and then perform dimension compression.

FIG. 7 is an explanatory diagram showing an example of high-order feature data. The high-level feature data includes high-level feature data 701 related to the non-chronological data 231, body type high-level feature data 702-1 related to the chronological data 232, blood pressure high-level feature data 702-2 related to the chronological data 232, and chronological data. blood glucose system high-level feature data 702-3 for 232 and liver function system high-level feature data 702-4 for time course data 232;

High-level feature data 701 related to the non-chronological data 231, body system high-level feature data 702-1 related to the chronological data 232, blood pressure high-level feature data 702-2 related to the chronological data 232, and blood glucose high-level feature data 702 related to the chronological data 232. -3, and liver function system high-order feature data 702-4 related to the chronological data 232 respectively include feature quantities 1, 2, . . . obtained by dimensional compression.

A vector Va1 is a set of values in the feature amount 1 column of the high-order feature data 701 related to the non-temporal data 231, and a vector Va2 is a set of values in the feature amount 2 column. A vector Vba1-1 is a set of values in the feature amount 1 column of the body type high-level feature data 702-1 related to the chronological data 232, and a vector Vb1-2 is a set of values in the feature amount 2 column.

A vector Vb2-1 is a set of values in the feature quantity 1 column of the blood pressure system high-level feature data 702-2 related to the chronological data 232, and a vector Vb2-2 is a set of values in the feature quantity 2 column. A vector Vb3-1 is a set of values in the feature amount 1 column of the blood glucose system high-level feature data 702-3 related to the chronological data 232, and a vector Vb3-2 is a set of values in the feature amount 2 column. A vector Vb4-1 is a set of values in the feature amount 1 column of the liver function system high-level feature data 702-4 related to the chronological data 232, and a vector Vb4-2 is a set of values in the feature amount 2 column.

Returning to FIG. 2 , the combining unit 205 combines a plurality of groups after dimension compression by the dimension compression unit 204 to generate combined high-order feature data 710 . The multiple groups after dimension compression are high-level feature data 702-1 to 702-n related to the temporal data 232 when the divided temporal feature data 600-1 to 600-n are dimensionally compressed. When the non-temporal data 231 and the divided temporal feature data 600-1 to 600-n are dimensionally compressed, the high-level feature data 701 for the non-temporal data 231 and the high-level feature data 702-1 to 702-n for the temporal data 232 be.

An example of coupling by the coupling unit 205 is shown with reference to FIG. , Vb1-1, Vb1-2, . . , Vb3-1, Vb3-2, . . . , Vb4-1, Vb4-2, . is generated as the combined high-order feature data 710 . , Vb1-1, Vb1-2, . . . , Vb3-1, Vb3-2, . summation.

Returning to FIG. 2, the analysis unit 206 uses the combination result (combined high-order feature data 710) by the combination unit 205 as an explanatory variable, and outputs a corresponding objective variable. For example, the analysis unit 206 performs insurance payment risk analysis, and specifically outputs the future probability of occurrence of death, hospitalization, surgery, hospital visits, etc. as objective variables. Specifically, for example, the analysis unit 206 performs data analysis using known techniques such as statistical techniques such as multiple regression analysis and machine learning techniques such as neural networks. Specifically, for example, the analysis unit 206 generates a learning model using a combination of the high-order feature vector Vall obtained from the existing notification information 300 and its analysis result as training data, and obtains from the new notification information 300 A new analysis result corresponding to the new notification information 300 is obtained by inputting the high-order feature vector Vall into the learning model.

<Screen example>
FIG. 8 is an explanatory diagram showing an input/output screen example 1. FIG. The input/output screen 800 is displayed on a display, which is an example of the output device 104 of the generation device 100 or a display of another computer that can communicate with the generation device 100 .

The input/output screen 800 includes a notification information read button 801, a domain knowledge read button 802, a feature extraction method selection pulldown 803, an analysis method selection pulldown 804, an analysis execution button 805, and an execution result display area 806. . A notification information read button 801 is a button that is pressed on the input device 103 . When the notice information read button 801 is pressed, the notice information 300 of the contract applicant stored in the storage device 102 is read.

The notification information 300 can be input by the input device 103 by pressing the notification information input button 807 to display the notification information input screen, instead of pressing the notification information read button 801 . FIG. 9 is an explanatory diagram showing an example of a notification information input screen. The notification information input screen 900 includes a health checkup result input area 901 and an interview result input area 902 . Values such as systolic blood pressure 323 , diastolic blood pressure 324 , fasting blood sugar 325 and the like can be set using the input device 103 in the health checkup result input area 901 . In the interview result input area 902 , the presence or absence of a smoking habit 331 , a drinking habit 332 , an exercise habit 333 , etc. can be set using the input device 103 .

Returning to FIG. 8 , a domain knowledge read button 802 is a button pressed on the input device 103 . When the domain knowledge read button 802 is pressed, the domain knowledge 400 stored in the storage device 102 is read. Alternatively, pressing the domain knowledge input button 808 can display a setting screen (not shown) for inputting domain knowledge. On the setting screen, it is possible to add, change, or delete the contents of the temporal data determination knowledge 410 , the temporal feature knowledge 420 , and the temporal feature division knowledge 430 by using the input device 103 .

A feature extraction method selection pull-down 803 is a button for displaying a pull-down list of a plurality of feature extraction methods on the input device 103 and selecting one of them. Multiple feature extraction techniques include, for example, well-known dimensionality reduction methods such as PCA and Stacked autoencoder as described above. For example, if PCA is selected, generator 100 will perform dimensionality compression with PCA.

An analysis method selection pull-down 804 is a button for displaying a pull-down display of a plurality of analysis methods on the input device 103 and for selecting one of them. The plurality of analysis methods include, for example, known methods such as statistical methods such as multiple regression analysis described above and machine learning methods such as neural networks. For example, when the multiple regression analysis is selected, the generation device 100 will perform data analysis using the multiple regression analysis.

An analysis execution button 805 is a button that is pressed on the input device 103 . When the analysis execution button is pressed, the generation device 100 loads the notification information 300 and the domain knowledge 400 from the storage device 102, performs dimensional compression using the method selected in the feature extraction method selection pull-down 803, and selects the analysis method selection pull-down. Data analysis is performed according to the technique selected at 804 . An execution result display area 806 is an area in which the execution result of the data analysis executed by pressing the analysis execution button 805 is displayed.

FIG. 10 is an explanatory diagram showing example 2 of the input/output screen. In FIG. 10 , an analysis result 1000 for each name ID 311 is displayed in the execution result display area 806 .

<Generation processing procedure example>
FIG. 11 is a flow chart showing an example of a generation processing procedure by the generation device 100. As shown in FIG. When the analysis execution button 805 is pressed, the generation device 100 reads the notification information 300 and the domain knowledge 400 (step S1101). The determination unit 201 determines whether the data is the temporal data 231 (step S1102). For the data determined to be the temporal data 232 (step S1102: Yes), the generating apparatus 100 causes the generating unit 202 to generate temporal feature data 500 (step S1103), and the dividing unit 203 divides the plurality of temporal feature data into a plurality of pieces. (step S1104).

For the data determined to be non-temporal data 231 (step S1102: No), the generating device 100 extracts high-level feature data 701 related to the non-temporal data 231 by the dimension compression unit 204 (step S1105). Similarly, the generation device 100 extracts high-order feature data 702-1 to 702-n regarding the temporal data 232 by the dimension compression unit 204 (steps S1106-1 to S1106-n).

Generating apparatus 100 combines high-level feature data 701 and 702-1 to 702-n by combining unit 205 to generate combined high-level feature data 710 (step S1107), and analysis unit 206 performs data analysis. Execute (step S1108) and display the analysis result 1000 in the execution result display area 806 of the input/output screen 800. FIG. This completes a series of generation processes.

As described above, according to the first embodiment, the chronological feature data 500 is grouped according to the chronological feature division knowledge 430, so generation of unnecessary feature amounts for analysis can be suppressed. As a result, it is possible to generate high-quality explanatory variables and improve the accuracy of data analysis. In addition, by suppressing the generation of feature amounts unnecessary for branching, the calculation cost can be reduced, and the calculation efficiency in data generation and data analysis can be improved.

Next, Example 2 will be described. In the second embodiment, differences from the first embodiment will be mainly described, so that the same components as those in the first embodiment are denoted by the same reference numerals, and the description thereof will be omitted.

FIG. 12 is a block diagram of a functional configuration example of the generation device 100 according to the second embodiment. The generation apparatus 100 according to the first embodiment executes the dimension compression processing by the dimension compression unit 204, the combination processing by the combination unit 205, and the analysis processing by the analysis unit 206 as independent processes. By applying the multimodal neural network 1200 in place of the unit 204, the combining unit 205, and the analyzing unit 206, the dimension compression processing by the dimension compression unit 204, the combining processing by the combining unit 205, and the analysis processing by the analyzing unit 206 can be performed continuously. execute

FIG. 13 is an explanatory diagram showing an example of a multimodal neural network 1200. As shown in FIG. FIG. 14 is an explanatory diagram showing an example of an input/output screen showing an analysis result 1400 by the multimodal neural network 1200. As shown in FIG.

The multimodal neural network 1200 first performs feature extraction on input vectors classified into a plurality of groups using neural networks f1, f2, . Next, the multimodal neural network 1200 connects the output vectors of the neural networks f1, f2, . Neural networks f1, f2, .

The learning of the multimodal neural network 1200 is supervised learning of input vectors input to the neural networks f1, f2, . The multimodal neural network 1200 can simultaneously learn three processes: feature extraction for each group, high-level feature extraction combining all groups, and analysis. Therefore, highly accurate analysis is possible depending on the design of the network structure and parameters. The operation of multimodal neural network 1200 is expressed, for example, by equation (1) shown in FIG.

h=g(f(Ua), ..., f(Ub1-1), f(Ub1-2), ..., f(Ub3-1), f(Ub3-2), ..., f(Ub4-1), f(Ub4-2),...)...(1)

Note that in equation (1), the vector Ua is a vector representation of the non-temporal data 231 . By giving the vector Ua to the function f, vectors Va1, Va2, . Also, by giving the vector Ub1-1, the vector Ub1-2, . . . to the function f, the vectors Vb1-1, Vba1-2, .

Also, by giving the vector Ub2-1, the vector Ub2-2, . . . to the function f, the vectors Vb2-1, Vba2-2, . Also, by giving the vector Ub3-1, the vector Ub3-2, . . . to the function f, the vectors Vb3-1, Vba3-2, . Also, by giving vector Ub4-1, vector Ub4-2, . . . to function f, vectors Vb4-1, Vba4-2, .

As described above, according to the second embodiment, since the dimension compression processing by the dimension compression unit 204, the combination processing by the combination unit 205, and the analysis processing by the analysis unit 206 are continuously executed, the generation processing and the analysis processing are performed at high speed. It is possible to improve the accuracy and accuracy of the measurement.

Further, in the first and second embodiments described above, the prediction of the insurance claim payment risk in the underwriting assessment of life insurance was explained as an example, but the present invention can also be applied to the financial analysis of companies. In this case, instead of the notification information 300, data described in the securities report or index data calculated from the data is used. Also, the temporal feature division knowledge 430 is, for example, information obtained by grouping the above data from five viewpoints of profitability, safety, activity, productivity, and growth.

Profitability is an item that indicates how much profit a company is making. Gross profit margin on sales, operating profit on sales, ordinary return on capital (ROA), return on equity (ROE) including. Safety is an item that indicates the ability of a company to pay for borrowings from banks, and includes current ratio, quick ratio, operating cash flow, investment cash flow, and the like. Activity is an item that indicates whether capital is used efficiently and sales are high, and includes total capital turnover, fixed asset turnover, inventory turnover, and the like.

Productivity is an item that indicates whether a company is using its employees and facilities efficiently, and includes sales value added ratio, labor share, and labor productivity. Growth potential is an item that indicates the future growth potential of a company, and includes sales growth rate, current profit growth rate, current net profit growth rate, and the like. As described above, the generation device 100 according to the first and second embodiments can be applied to various data analyses.

It should be noted that the present invention is not limited to the embodiments described above, but includes various modifications and equivalent configurations within the scope of the appended claims. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to those having all the described configurations. Also, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Moreover, the configuration of another embodiment may be added to the configuration of one embodiment. Moreover, other configurations may be added, deleted, or replaced with respect to a part of the configuration of each embodiment.

Further, each configuration, function, processing unit, processing means, etc. described above may be realized by hardware, for example, by designing a part or all of them with an integrated circuit, and the processor 101 performs each function. It may be realized by software by interpreting and executing a program to be realized.

Information such as programs, tables, files, etc. that realize each function is stored in storage devices such as memory, hard disk, SSD (Solid State Drive), or IC (Integrated Circuit) card, SD card, DVD (Digital Versatile Disc) recording Can be stored on media.

In addition, the control lines and information lines indicate those considered necessary for explanation, and do not necessarily indicate all the control lines and information lines necessary for mounting. In practice, it can be considered that almost all configurations are interconnected.

100 generation device 101 processor 102 storage device 201 determination unit 202 generation unit 203 division unit 204 dimension compression unit 205 combination unit 206 analysis unit 300 notification information 400 domain knowledge 500 temporal feature data 600 divided temporal feature data 700 high-level feature data 710 connection height Next Feature Data 1200 Multimodal Neural Network

Claims (7)

A generating device having a processor that executes a program and a storage device that stores the program,
Accessible to temporal feature information indicating temporal features obtained from the temporal data, and grouping information defining a plurality of groups to which the temporal data should belong,
The processor
a generation process of generating a plurality of pieces of temporal feature data indicating the temporal feature from the temporal data to be analyzed based on the temporal feature information;
a division process for dividing the plurality of temporal feature data generated by the generation process into the plurality of groups based on the grouping information;
Dimensional compression processing for dimensionally compressing each of the plurality of groups divided by the division processing, and dimensionally compressing non-temporal data to be analyzed ;
A generating device characterized by executing The generator of claim 1, comprising:
Accessible to determination information for determining which of the chronological data and the non-chronological data corresponds,
The processor
executing determination processing for determining whether the data to be analyzed corresponds to the temporal data or the non-temporal data based on the determination information;
In the generation process, the processor treats the analysis target data determined to be the temporal data by the determination process as the analysis target temporal data, and based on the temporal feature information, the analysis target temporal data is: generating the plurality of temporal feature data;
In the dimension compression processing, the processor treats the analysis target data determined to be the non-temporal data by the determination processing as the analysis target non-temporal data, and dimensionally compresses the analysis target non-temporal data.
A generating device characterized by: A generating device according to claim 2, comprising:
A generating device that performs a combining process for combining non-temporal data to be analyzed after the dimension compression by the dimension compression process and a plurality of groups after the dimension compression. A generating device according to claim 3, comprising:
The processor
A generation apparatus characterized by executing an analysis process for outputting a corresponding objective variable using a combination result obtained by the combination process as an explanatory variable. 5. A generating device according to claim 4,
A generating device, wherein the dimensionality compression processing, the connection processing, and the analysis processing are performed by a multimodal neural network. A generation method executed by a generation device having a processor that executes a program and a storage device that stores the program,
The generation device is capable of accessing temporal characteristic information indicating temporal characteristics of temporal data and grouping information defining a plurality of groups to which the temporal data should belong,
The processor
a generation process of generating a plurality of pieces of temporal feature data indicating the temporal feature from the temporal data to be analyzed based on the temporal feature information;
a division process for dividing the plurality of temporal feature data generated by the generation process into the plurality of groups based on the grouping information;
Dimensional compression processing for dimensionally compressing each of the plurality of groups divided by the division processing, and dimensionally compressing non-temporal data to be analyzed;
A generation method characterized by executing A generator program for causing a processor to perform data generation, comprising:
The processor is capable of accessing temporal characteristic information indicating temporal characteristics of the temporal data and grouping information defining a plurality of groups to which the temporal data should belong,
to the processor;
a generation process of generating a plurality of pieces of temporal feature data indicating the temporal feature from the temporal data to be analyzed based on the temporal feature information;
a division process for dividing the plurality of temporal feature data generated by the generation process into the plurality of groups based on the grouping information;
Dimensional compression processing for dimensionally compressing each of the plurality of groups divided by the division processing, and dimensionally compressing non-temporal data to be analyzed;
A generation program characterized by executing

Images