Time discount rate estimation device, machine learning method, and program

Abstract

The purpose of the present disclosure is to highly accurately estimate a temporal discounting rate of a user, without using a measurement method that uses a survey.　To achieve this aim, the present disclosure is a temporal discounting rate estimation device that estimates a temporal discounting rate in a training phase, said temporal discounting rate estimation device having a model training unit that: after normalizing values related to a plurality of behavioral features of a user, calculates the difference between a value obtained by multiplying said normalized values by a model parameter indicating coefficients serving as weights, and a temporal discounting rate that is correct answer data based on a response by the user; and carries out machine learning of the model parameter such that said difference becomes smaller.

Description

[Technical Field]

[0001] The present disclosure relates to a technology for analyzing a time discount rate, and more particularly to a technology for automatically estimating a user's time discount rate with high accuracy from the user's daily behavior record. [Background technology]

[0002] Quantifying human personality is an important element in understanding human beings, as tackled in fields such as psychology and economics. Traditionally, economics treats human personality as a function of three factors: "time discount rate," "risk aversion," and "reciprocity." Of these, the time discount rate is an index that focuses on a person's "aversion to waiting." In particular, it focuses on the human tendency to discount a reward over time, based on how much the future value of a reward (delayed reward) is subjectively perceived as lower than its present value (immediate reward). The time discount rate is treated as a parameter of the discount decay function (e.g., exponential or bipolar). Statistical research on time discount rate, primarily in behavioral economics, has been conducted in conjunction with human attributes, revealing that it is an important factor affecting a wide range of life domains. For example, a high time discount rate (aversion to waiting) is correlated with higher rates of debt, obesity, and smoking (Non-Patent Document 1). By clarifying each individual's time discount rate, individuals and others can quantitatively understand how much they can tolerate waiting in their daily lives, and this can lead to support in decision-making to improve their lives.

[0003] To date, methods for measuring time discount rates using questionnaires have been developed. Two main measurement methods have been proposed and are used in various statistical surveys (Non-Patent Document 2). One method is a measurement method using a multiple-choice questionnaire. For example, a question such as "Would you rather receive X yen today (option A) or Y yen in seven days (option B)?" is prepared, with Y yen being higher than X yen in both cases, and respondents are asked to choose from these options. In this case, since option B is higher, the question designer creates an "annual interest rate" similar to that of a bank deposit. Focusing on questions where the response switched from option A (receive today) to option B (receive in seven days), the respondent's discount rate is assumed to lie between the annual interest rate in the question where the switch occurred and the annual interest rate in the previous question. The average of these two annual interest rates is used as the time discount rate. Another method is to measure using a "fill-in questionnaire," which consists of a single question, for example, "If you were to receive X yen today and if you were to receive it seven days later, please enter the amount Y that would have the same value." The annual interest rate calculated from the difference between the entered amount Y yen and X yen is then used as the time discount rate. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Mischel, Walter et al. "'Willpower' over the life span: decomposing self-regulation." Social cognitive and affective neuroscience, Vol. 6, No. 2 (2011): 252-6. [Non-patent document 2] Hardisty, David J. et al. "How to measure time preferences: An experimental comparison of three methods." Judgment and Decision Making Vol. 8 (2013): 236-249. Summary of the Invention [Problem to be solved by the invention]

[0005] However, while measurement methods using written questionnaires place a small burden on respondents because they only need to answer a single question, allowing them to enter any amount they like, the responses from different groups of respondents tend to diverge, making it difficult to measure an accurate time discount rate and making it difficult to compare the results with the attribute information of the respondent groups.

[0006] Furthermore, while measurement methods using multiple-choice questionnaires are less likely to result in divergent responses because respondents' time discount rates are assigned based on an annual interest rate determined in advance by the question designer, they require respondents to answer multiple questions. Furthermore, to prevent respondents from responding randomly (for example, by continuing to choose option B without thinking), the order of the questions is randomly shuffled according to the annual interest rate. Therefore, when there are multiple points where option A switches to option B, many respondents are treated as having invalid answers, and in many cases, proper measurement is not possible. Furthermore, requiring respondents to answer multiple questions places a heavy burden on respondents, and because the survey is conducted over a medium- to long-term time interval, it is difficult to detect changes.

[0007] The present invention has been made in consideration of the above points, and has as its object to estimate a user's time discount rate with high accuracy without relying on a measurement method using a questionnaire. [Means for solving the problem]

[0008] In order to achieve the above object, the invention of claim 1 is a time discount rate estimation device that estimates a time discount rate in a learning phase, the device comprising: a model learning unit that calculates an error between a value obtained by standardizing each value related to a plurality of behavioral features of a predetermined user and then multiplying the standardized value by a model parameter indicating each coefficient used as a weighting coefficient, and a time discount rate that is correct answer data based on an answer by the predetermined user; and machine learning the model parameter so as to reduce the error. a behavior data preprocessing unit that generates preprocessed behavior data by calculating summary statistics for each attribute value and calculating the duration of each behavior based on behavior amount data that numerically indicates the behavior of the predetermined user over time and behavior date and time data that indicates the type of behavior of the predetermined user and the date and time of occurrence of the behavior related to the type; a behavior statistics calculation unit that generates behavior statistical feature data by calculating an average value for each behavior amount based on the preprocessed behavior data; a behavior date and time preprocessing unit that generates preprocessed behavior date and time data that includes a behavior relationship that indicates the relationship between a current behavior and a next behavior and indicates the start date and time of another type of behavior that occurred from the current behavior to the start of a next behavior of the same type based on the behavior date and time data; a behavior transition calculation unit that generates behavior transition feature data by calculating a transition probability between the current behavior and the next behavior based on the preprocessed behavior date and time data; and a behavior statistical calculation unit that performs labeling for division at a predetermined date based on date and time information in the preprocessed behavior data, divides the behavior statistical feature data based on the label, and calculates the absolute value of the difference amount of behavior statistics for each of the divided data. a behavioral variance calculation unit that generates variance data relating to the behavioral statistics, labels the preprocessed behavior date and time data for division by predetermined dates based on date and time information in the preprocessed behavior date and time data, divides the behavior transition feature data based on the labels, calculates absolute values ​​of variances of behavior statistics for each divided data to generate variance data relating to behavior transitions, and combines the variance data relating to the behavior statistics and the behavior transition feature data for the same user to generate behavioral variance feature data; a behavioral feature correlation calculation unit that calculates a time discount rate related to the time discount rate data as the correct answer data and a Pearson's correlation coefficient for the behavioral statistics feature data, the behavior transition feature data, and the behavior variance feature data for the specified user, and calculates a test statistic for the correlation coefficient; and a feature selection unit that selects, from each of the behavioral features related to the behavioral statistics feature data, the behavior transition feature data, and the behavior variance feature data, multiple behavioral features that are correlated with the time discount rate related to the correct answer data and do not have a similar tendency, and combines them for each user to generate various behavioral feature data,standardizing the values ​​of the plurality of behavioral features selected by the feature selection unit; A time discount rate estimation device. [Effects of the Invention]

[0009] As described above, the present invention has the effect of enabling a user's time discount rate to be estimated with high accuracy without relying on a measurement method using a questionnaire. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 2 is a functional configuration diagram of a time discount factor estimation device in a learning phase according to an embodiment. [Figure 2] FIG. 2 is a functional configuration diagram of a time discount rate estimation device in an estimation phase according to an embodiment. [Figure 3] FIG. 2 is a hardware configuration diagram of a time discount rate analysis device according to an embodiment. [Figure 4] 1A and 1B show tables constituting the behavior data DB, in which FIG. 1A is a conceptual diagram of an activity amount management table, and FIG. 1B is a conceptual diagram of an activity date and time management table. [Figure 5] FIG. 2 is a conceptual diagram of a table constituting a time discount rate data DB. [Figure 6] FIG. 2 is a conceptual diagram of a table constituting an evaluation value storage DB. [Figure 7] 10 is a flowchart showing an outline of a process for estimating a time discount rate in a learning phase. [Figure 8] 10 is a flowchart showing an outline of a process for estimating a time discount rate in the estimation phase. [Figure 9] 10 is a flowchart showing the processing of a behavioral data preprocessing unit. [Figure 10] 10A and 10B are conceptual diagrams showing examples of output from the behavioral data preprocessing unit. [Figure 11] 10 is a flowchart showing the processing of a behavioral statistics calculation unit. [Figure 12] FIG. 10 is a conceptual diagram showing an example of output from a behavioral statistics calculation unit. [Figure 13] 10 is a flowchart showing the processing of an action date and time preprocessing unit. [Figure 14] FIG. 10 is a conceptual diagram illustrating an example of output from an action date and time preprocessing unit. [Figure 15]10 is a flowchart showing the processing of a behavior transition calculation unit. [Figure 16] FIG. 10 is a conceptual diagram illustrating an example of an output from a behavior transition calculation unit. [Figure 17] 10 is a flowchart showing the processing of a behavioral difference amount calculation unit. [Figure 18] FIG. 10 is a conceptual diagram showing an example of output from a behavioral difference amount calculation unit. [Figure 19] 10 is a flowchart showing the processing of a behavioral feature correlation calculation unit. [Figure 20] FIG. 10 is a conceptual diagram showing an example of output from a behavioral feature correlation calculation unit. [Figure 21] 10 is a flowchart showing the processing of a feature selection unit. [Figure 22] 10 is a flowchart showing the processing of the model learning and selection unit. [Figure 23] 10 is a flowchart showing the processing of a time discount factor estimation unit. DETAILED DESCRIPTION OF THE INVENTION

[0011] [Outline of the embodiment] In recent years, wearable devices have become increasingly common among individuals, making it easier to observe and digitize daily behavior. Behavior is observed as a result of decision-making. Since time discounting rates have been confirmed to correlate with lifestyle behavioral outcomes such as obesity and smoking, it is thought that analyzing more detailed patterns of daily behavior (e.g., sleep, diet, exercise) can estimate the time discounting rate underlying decision-making. Furthermore, estimating time discounting rates from this automatically measured behavioral data can reduce the burden of individuals completing surveys and reveal time discounting rates at finer time granularities (e.g., weekly or monthly), potentially supporting self-understanding and decision-making over time.

[0012] The time discount factor estimation device 1 of this embodiment estimates an individual's time discount factor with high accuracy from behavioral data automatically observed by a wearable device or the like, without relying on a measurement method using a questionnaire.

[0013] [Hardware configuration of the time discount factor analysis device] Next, the hardware configuration of the time discount factor analysis device 1 will be described with reference to Fig. 3. Fig. 3 is a diagram showing the hardware configuration of the time discount factor analysis device according to the embodiment.

[0014] 3, time discount factor analysis device 1 includes processor 101, memory 102, auxiliary storage device 103, connection device 104, communication device 105, and drive device 106. The hardware components that make up time discount factor analysis device 1 are connected to each other via bus 107.

[0015] Processor 101 serves as a control unit that controls the entire time discount factor analysis device 1, and includes various computing devices such as a CPU (Central Processing Unit). Processor 101 reads and executes various programs onto memory 102. Processor 101 may also include a GPGPU (General-purpose computing on graphics processing units).

[0016] The memory 102 has a main storage device such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The processor 101 and the memory 102 form a so-called computer, and the processor 101 executes various programs read onto the memory 102, thereby enabling the computer to realize various functions.

[0017] The auxiliary storage device 103 stores various programs and various information used when the processor 101 executes the various programs.

[0018] The connection device 104 is a connection device that connects an external device (for example, the display device 110, the operation device 111) and the time discount factor analysis device 1.

[0019] The communication device 105 is a communication device for transmitting and receiving various types of information to and from other devices.

[0020] The drive device 106 is a device for loading a (non-transitory) recording medium 130. The recording medium 130 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM (Compact Disc Read-Only Memory), a flexible disk, or a magneto-optical disk. The recording medium 130 may also include semiconductor memory that records information electrically, such as a ROM (Read Only Memory) or flash memory.

[0021] The various programs to be installed in the auxiliary storage device 103 are installed, for example, by setting the distributed recording medium 130 in the drive device 106 and reading the various programs recorded on the recording medium 130 by the drive device 106. Alternatively, the various programs to be installed in the auxiliary storage device 103 may be installed by being downloaded from a network via the communication device 105.

[0022] [Functional configuration of the time discount factor analysis device] An embodiment of the present invention will be described below. Fig. 1 is a functional configuration diagram of a time discount factor estimation device in a learning phase of the embodiment. Fig. 2 is a functional configuration diagram of a time discount factor estimation device in an estimation phase of the embodiment.

[0023] 1, the time discount factor estimation device 1 in the learning phase includes a behavioral data preprocessing unit 11, a behavioral statistics calculation unit 12, a behavioral date and time preprocessing unit 13, a behavioral transition calculation unit 14, a behavioral variance calculation unit 15, a behavioral feature correlation calculation unit 16, a feature selection unit 17, and a model learning and selection unit 18. Each of these units is a function realized by instructions from the processor 101 in FIG. 3 based on a program.

[0024] Furthermore, the time discount factor estimation device 1 in the learning phase has a behavioral data DB (Data Base) 21, a time discount factor data DB 22, an evaluation value storage DB 23, and a time discount factor estimation model DB 24. Each of these DBs is constructed in memory 102 or auxiliary storage device 203, which will be described later. The time discount factor estimation device 1 in the learning phase uses information from each DB to output a trained time discount factor estimation model. Note that hereinafter, the time discount factor model may be simply referred to as the "model."

[0025] 2, the time discount factor estimation device 1 in the estimation phase includes a behavioral data preprocessing unit 11, a behavioral statistics calculation unit 12, a behavioral date and time preprocessing unit 13, a behavioral transition calculation unit 14, a behavioral variance calculation unit 15, and a time discount factor estimation unit 19. Each of these units is a function realized by instructions from a processor 101 (shown in FIG. 3, described below) based on a program.

[0026] Furthermore, in the estimation phase, the time discount factor estimation device 1 has a time discount factor estimation model DB 24 and an unknown user behavior data DB 25. Each of these DBs is constructed in memory 102 or auxiliary storage device 203. In the learning phase, the time discount factor estimation device 1 uses information from each DB to output a trained time discount estimation model.

[0027] <Behavioral Data DB> FIG. 4 shows tables that make up the behavior data DB, where (a) is a conceptual diagram of an activity amount management table and (b) is a conceptual diagram of an activity date and time management table. The behavior data BD21 stores behaviors that are automatically recorded by a wearable device or recorded by the user, along with a user ID, date and time, and the amount of the behavior, or a character string representing the user ID, date and time, and the type (content) of the behavior. The type of behavior may be stored in the behavior data DB21 to the extent that it can be collected by the system administrator. The user ID is an example of user identification information, and may be assigned a symbol or number that uniquely identifies the user. Specifically, the table is configured as follows:

[0028] (Action amount management table) As shown in Figure 4(a), the activity amount management table manages activity amount data that associates the activity date and time, number of steps, heart rate, etc. of the user identified by the user ID with the user ID.

[0029] (Action date and time management table) Also, as shown in Figure 4(b), the activity date and time management table manages activity date and time data that associates the activity date and time of the user identified by the user ID and the type (content) of the activity with the user ID.

[0030] <Time discount rate data DB> 5 is a conceptual diagram of a table constituting the time discount rate data DB 22. In the time discount rate data DB 22, the time discount rate is managed for each user ID.

[0031] <Evaluation value storage DB> 6 is a conceptual diagram of tables constituting the evaluation value storage DB 23. In the evaluation value storage DB 23, the mean absolute error and mean square error of each machine learning model are managed for each type of model.

[0032] <Time discount rate estimation model DB> The time discount rate estimation model DB 24 is a DB that manages time discount rate estimation models for machine learning, and a description thereof will be omitted here.

[0033] <Behavioral data database of unknown users> The unknown user behavior data DB 25 has the same data structure as the behavior data DB (see FIGS. 4(a) and 4(b)), and manages the unknown user behavior data as input data used for estimating the time discount rate.

[0034] <Each function configuration> Next, each functional configuration of the time discount factor estimation device 1 in the learning phase will be described.

[0035] The behavioral data preprocessing unit 11 generates preprocessed behavioral data by calculating summary statistics for each attribute value and the duration of a specified behavior based on behavioral amount data, which is behavioral data indicating the behavior of a specified user numerically over time, and behavioral date and time data, which is indicating the type of behavior of the specified user and the date and time of occurrence of the behavior related to that type.

[0036] The behavioral statistics calculation unit 12 generates behavioral statistical feature data by calculating an average value for each behavior amount based on the preprocessed behavioral data.

[0037] The behavior date and time preprocessing unit 13 generates preprocessed behavior date and time data based on the behavior date and time data, which includes a behavior relationship indicating the relationship between the current behavior and the next behavior and indicates the latest start dates and times of all types of behaviors viewed from the current behavior.

[0038] The behavior transition calculation unit 14 generates behavior transition feature data by calculating the transition probability between the current behavior and the next behavior based on the preprocessed behavior date and time data.

[0039] The behavioral variance calculation unit 15 performs labeling for division at predetermined dates based on date and time information in the preprocessed behavioral data, divides the behavioral statistical feature data based on the labels, and generates variance data related to behavioral statistics by calculating the absolute value of the variance of the behavioral statistics for each divided data, and also performs labeling for division at predetermined dates based on date and time information in the preprocessed behavioral date and time data, divides the behavioral transition feature data based on the labels, and generates variance data related to behavioral transitions by calculating the absolute value of the variance of the behavioral statistics for each divided data, and generates behavioral variance feature data by combining the variance data related to behavior statistics and the behavioral transition feature data for the same predetermined user.

[0040] The behavioral feature correlation calculation unit 16 calculates a Pearson correlation coefficient between the time discount rate related to the time discount rate data as correct data and the behavioral statistical feature data, behavioral transition feature data, and behavioral variance feature data related to the same user, and also calculates a test statistic for this correlation coefficient.

[0041] The feature selection unit 17 selects, from among the behavioral features related to the behavioral statistical feature data, the behavioral transition feature data, and the behavioral difference feature data, multiple behavioral features that are correlated with the time discount rate related to the correct answer data and do not have similar trends, and generates various behavioral feature data by combining them for the same specified user.

[0042] The model learning and selection unit 18 standardizes the values ​​of the multiple behavioral features selected by the feature selection unit 17, and then, using a supervised machine learning method, calculates the error between the value obtained by multiplying each standardized value by a model parameter indicating each weighting coefficient and the time discount rate, which is the correct data, and learns the model parameters (time discount rate estimation model) so as to reduce the error.

[0043] The time discount rate estimation unit 19 uses machine-learned model parameters (time discount rate estimation model) to estimate the time discount rate of a specific individual user from the input data, which is behavioral data of a specific unknown individual user, and outputs the resulting time discount rate of the individual.

[0044] Each part will be explained in detail later.

[0045] [Processing or Operation of the Embodiment] Next, the processing or operation of this embodiment will be described in outline and then in detail, with the processing being divided into a learning phase and an estimation phase.

[0046] <Processing Overview> (Outline of the learning phase process) FIG. 7 is a flowchart showing an outline of the process of estimating the time discount rate in the learning phase.

[0047] First, the behavioral data preprocessing unit 11 receives and processes the behavioral data of each person from the behavioral data DB 21 (S100). The details of this processing will be described later. As shown in Fig. 4, the behavioral data is roughly divided into behavior amount data (number of steps, heart rate, conversation time, etc.) in which the behavior is recorded numerically over time (see Fig. 4(a)), and behavior date and time data in which the type of behavior and the date and time of its occurrence are recorded (see Fig. 4(b)), and each is recorded together with the user ID.

[0048] The behavioral statistics calculation unit 12 receives and processes the preprocessed behavioral data from the behavioral data preprocessing unit 11 (S110). Details of the processing will be described later. Figure 12 shows an example of data obtained as output data from the behavioral statistics calculation unit 12.

[0049] The activity date and time preprocessing unit 13 receives the preprocessed activity data from the activity data preprocessing unit 11 and receives and processes data from the activity data DB 21 (S120). Details of the processing will be described later. Figure 14 shows an example of data obtained as output data from the activity date and time preprocessing unit 13.

[0050] The behavior transition calculation unit 14 receives and processes the behavior date and time data from the behavior date and time preprocessing unit 13 (S130). The details of the processing will be described later. Fig. 16 shows an example of data obtained as output data from the behavior transition calculation unit 14.

[0051] The behavioral difference amount calculation unit 15 receives and processes the behavioral statistical feature data from the behavioral statistics calculation unit 12 and the behavioral transition feature data from the behavioral transition calculation unit 14 (S140). Details of the processing will be described later. Fig. 18 shows an example of data obtained as output data from the behavioral difference amount calculation unit 15.

[0052] The behavioral feature correlation calculation unit 16 receives the behavioral statistical feature data, behavioral transition feature data, and behavioral difference feature data from the behavioral variance amount calculation unit 15, and receives time discount rate data as machine learning correct answer data from the time discount rate data DB 22, and processes them (S150). Details of the processing will be described later. Fig. 20 shows an example of data obtained as output data from the behavioral feature correlation calculation unit 16.

[0053] The feature selection unit 17 receives and processes the correlation coefficients and significance probability data for various feature data from the behavioral feature correlation calculation unit 16 (S160). The details of the processing will be described later.

[0054] The model learning and selection unit 18 receives the time discount rate from the time discount rate data DB 22, receives various feature data from the feature selection unit 17, learns the model and verifies its accuracy, and outputs the trained model with the highest accuracy to the time discount estimation model DB 24 (S170). Details of this process will be described later.

[0055] (Outline of the estimation phase process) FIG. 8 is a flowchart showing an outline of the processing for estimating the time discount rate in the estimation phase.

[0056] First, the behavior data preprocessing unit 11 receives and processes data from the behavior data DB 25 of unknown users as input data (S200).

[0057] The behavioral statistics calculation unit 12 receives and processes the preprocessed behavioral data from the behavioral data preprocessing unit 11 (S210).

[0058] The activity date and time preprocessing unit 13 receives the preprocessed activity data from the activity data preprocessing unit 11, and receives and processes the data from the activity data DB 21 (S220).

[0059] The behavior transition calculation unit 14 receives and processes the preprocessed behavior date and time data from the behavior date and time preprocessing unit 13 (S230).

[0060] The behavior difference amount calculation unit 15 receives the behavior statistical characteristic data from the behavior statistics calculation unit 12 and the behavior transition characteristic data from the behavior transition calculation unit 14, and processes them (S240).

[0061] The time discount factor estimation unit 19 receives the trained model from the time discount estimation model DB 24 and various feature data from the behavioral variance calculation unit 15, and outputs the time discount factor as the estimation result (S250). The details of this process will be described later.

[0062] <Detailed processing> Next, the detailed processing of the learning phase will be described.

[0063] (Details of the behavioral data preprocessing section) The detailed processing of the behavior data pre-processing unit 11 will be described with reference to Fig. 9. Fig. 9 is a flowchart showing the processing of the behavior data pre-processing unit.

[0064] First, the behavioral data preprocessing unit 11 receives, as examples of behavioral data, behavioral amount data (behavior amount management table) as shown in Figure 4(a) and behavioral date and time data (behavior date and time management table) as shown in Figure 4(b) from the behavioral data DB21 in the learning phase, or from the behavioral data DB25 of an unknown user as input data in the estimation phase (S300).

[0065] The behavioral data preprocessing unit 11 aggregates the behavioral amount data by user ID and date, and calculates summary statistics for each attribute value (S310). The statistics calculated here include the sum, average, standard deviation, maximum, and minimum values ​​of the aggregated values. The system administrator can also specify the statistics to be applied for each attribute. The example in Figure 10 is an example where the "sum" is specified for the behavioral amount "number of steps," and the average, standard deviation, maximum, and minimum values ​​are specified for the "heart rate."

[0066] The behavioral data preprocessing unit 11 aggregates data for each user ID and date in the behavior date and time table, and calculates the total time for behaviors for which the "start" and "end" of the behavior are explicitly specified and the duration can be calculated. For example, the example in Fig. 10 shows that the duration is calculated from the difference between the dates and times of the behaviors "start of sleep" and "end of sleep," "start of work" and "end of work," and "start of exercise" and "end of exercise," and further, the total time is calculated for durations observed multiple times in a day.

[0067] The behavioral data preprocessing unit 11 passes the converted preprocessed behavioral data (see FIGS. 10(a) and 10(b)) to the behavioral statistics calculation unit (S330).

[0068] (Detailed processing of behavioral statistics calculation section) Detailed processing of the behavior statistic calculation unit 12 will be described with reference to Fig. 11. Fig. 11 is a flowchart showing processing of the behavior statistic calculation unit.

[0069] First, the behavioral statistics calculation unit 12 receives preprocessed behavioral data from the behavioral data preprocessing unit 11 (S400).

[0070] The behavioral statistics calculation unit 12 aggregates data for each user ID and calculates the average value for each behavior amount (that is, for each column) (S410).

[0071] The behavioral statistics calculation unit 12 passes the converted behavioral statistical feature data and the preprocessed behavioral data (see FIG. 12) to the behavioral difference calculation unit (S420).

[0072] (Detailed processing of the action date and time preprocessor) The detailed processing of the activity date and time preprocessing unit 13 will be described with reference to Fig. 13. Fig. 13 is a flowchart showing the processing of the activity date and time preprocessing unit.

[0073] The behavior date and time preprocessing unit 13 receives behavior date and time data (behavior date and time management table) as shown in Fig. 4(b) from the behavior data DB 21 in the learning phase, or from the behavior data DB 25 of the unknown user as input data in the estimation phase (S500). Note that the behavior amount data in Fig. 4(a) is not required here.

[0074] The activity date and time preprocessing unit 13 aggregates the activity date and time data for each user ID, and sorts the data in ascending order by date and time (S510).

[0075] The behavior date and time preprocessing unit 13 stores data on the behavior recorded at each date and time as "current behavior" and the behavior recorded at the next date and time as "next behavior" in the memory 102, etc. At this time, it stores the type of behavior recorded as "current behavior" (S520).

[0076] The behavior date and time preprocessing unit 13 records (S523) the date and time of observation after the date and time of each current behavior for all behaviors confirmed in the processes (S510, S520) for all "current behaviors." The output example of Fig. 14 shows data recorded in the memory 102, etc., including the dates and times of "sleep start," "sleep end," and "weight measurement."

[0077] The behavior date and time preprocessing unit 13 passes the converted preprocessed behavior date and time data (see FIG. 14) to the behavior transition calculation unit 14 (S540).

[0078] (Detailed processing of the behavior transition calculation section) The detailed processing of the behavior transition calculation unit 14 will be described with reference to Fig. 15. Fig. 15 is a flowchart showing the processing of the behavior transition calculation unit.

[0079] First, the behavior transition calculation unit 14 receives preprocessed behavior date and time data from the behavior date and time preprocessing unit 13 (S600).

[0080] The behavior transition calculation unit 14 calculates the transition probability between the behaviors of each user (S610). Specifically, the behavior transition calculation unit 14 aggregates data for each user ID and "current behavior," counts the number of pieces of data, and regards the counted number as the observed number of the current behavior. Next, the behavior transition calculation unit 14 aggregates data for each user ID, "current behavior," and "next behavior," counts the number of pieces of data, and regards this as the simultaneous observed number of the current behavior and the next behavior. For each behavior type, the behavior transition calculation unit 14 divides the simultaneous observed number of the current behavior and the next behavior by the observed number of the current behavior, and stores the value in the memory 102 as the transition probability between each current behavior and each next behavior.

[0081] The behavior transition calculation unit 14 calculates the average transition time between the behaviors of each user (S620). Specifically, the behavior transition calculation unit 14 aggregates data for each user ID and "current behavior," calculates the difference between the next observed date and time for each behavior, and stores the average value of the differences as the average transition time between each current behavior and each next behavior.

[0082] The behavior transition calculation unit 14 calculates the standard deviation of transition time between the behaviors of each user (S630). Specifically, the behavior transition calculation unit 14 aggregates data for each user ID and "current behavior," calculates the difference between the date and time of the next observation for each behavior, and stores the standard deviation of the difference in the memory 102 as the standard deviation of transition time between each current behavior and each next behavior.

[0083] The behavior transition calculation unit 14 passes the preprocessed behavior date and time data and the behavior transition feature data (see FIG. 16) calculated by the behavior transition calculation unit 14 to the behavior difference amount calculation unit 15 (S640).

[0084] (Detailed processing of the behavioral difference calculation unit) Detailed processing of the behavioral difference amount calculation unit 15 will be described with reference to Fig. 17. Fig. 17 is a flowchart showing processing of the behavioral difference amount calculation unit.

[0085] First, the behavioral difference calculation unit 15 receives preprocessed behavioral data and behavioral statistical feature data from the behavioral statistics calculation unit 12, and receives preprocessed behavioral date and time data and behavioral transition feature data from the behavior transition calculation unit (S700).

[0086] The behavioral variance calculation unit 15 labels the preprocessed behavioral data and the dates of the behavioral date and time data (S710). Here, the system administrator selects and assigns two types of label patterns in consideration of human life in the real world. The first is the labels of "weekdays" (Monday to Friday) and "holidays" (Saturday and Sunday) obtained by referencing a calendar from the date information. The second is the label of "work present" if "work starts" is included in the current behavior when the preprocessed behavioral date and time data is referenced and aggregated by date, and the label of "work" is used if this behavior does not include "work".

[0087] The behavioral difference calculation unit 15 refers to the two types of labels obtained in the process (S710) and calculates the difference between the behavioral statistics (S720). Specifically, the behavioral difference calculation unit 15 extracts data labeled "weekday" from the preprocessed behavioral data, and invokes and executes the process (S410) of the behavioral statistics calculation unit 12. Similarly, the behavioral difference calculation unit 15 extracts data labeled "holiday" and invokes and processes the process (S410) of the behavioral statistics calculation unit 12. The two output results are combined by user ID, and the absolute value of the difference between the weekday behavior statistics and the holiday behavior statistics for the same statistics is obtained. When the labels "work" and "no work" are selected, the same process as above is executed by referring to the labels.

[0088] The behavioral difference calculation unit 15 refers to the two types of labels obtained in the process (S710) and calculates the difference in the behavioral transition features (S730). Specifically, the behavioral difference calculation unit 15 extracts data labeled "weekday" from the preprocessed behavioral date and time data and invokes the processes (S610, S620, S630) of the behavioral transition calculation unit 14 for processing. Similarly, the behavioral difference calculation unit 15 extracts data labeled "holiday" and invokes the processes (S610, S620, S630) of the behavioral transition calculation unit 14 for processing. The behavioral difference calculation unit 15 combines the two output results by user ID and obtains the absolute value of the difference between the weekday behavior statistics and the holiday behavior statistics for the same statistics. Even when the labels "work" and "no work" are selected, the behavioral difference calculation unit 15 refers to the labels and performs the same process as above.

[0089] The behavioral difference amount calculation unit 15 combines the difference amount data related to behavior statistics and the difference amount data related to behavioral transitions by the user ID (S740). The data obtained as a result is output from the behavioral difference amount calculation unit 15 as behavioral difference amount feature data.

[0090] The behavioral difference amount calculation unit 15 passes the behavioral statistical feature data, behavioral transition feature data, and behavioral difference amount feature data (see Figure 18) to the behavioral feature correlation calculation unit 16 in the learning phase, and to the time discount rate estimation unit 19 in the estimation phase (S750).

[0091] (Detailed processing of behavioral feature correlation calculation section) Detailed processing of the behavioral feature correlation calculation unit 16 will be described with reference to Fig. 19. Fig. 19 is a flowchart showing processing of the behavioral feature correlation calculation unit.

[0092] First, the behavioral feature correlation calculation unit 16 receives the behavior statistical feature data, the behavior transition feature data, and the behavioral difference amount feature data from the behavioral difference amount calculation unit 15 (S800).

[0093] Behavioral feature correlation calculation unit 16 receives time discount rate data as correct answer data from time discount rate data DB 22 (S810).

[0094] The behavioral feature correlation calculation unit 16 links each feature data and the time discount rate data by the user ID (S820). Here, the user ID is used to link the data because the person who answered the questionnaire or the like that is the source of the correct answer data must be the same person who performed the behavior related to the behavioral data.

[0095] The behavioral feature correlation calculation unit 16 calculates Pearson's correlation coefficient for the time discount rate assigned to each user and each feature data, and stores the calculated coefficient in the memory 102 in the format shown in FIG. 20 (S830). This correlation indicates, for example, that the higher the time discount rate of a person, the higher the smoking rate. FIG. 20 shows, for example, that the correlation coefficient between the behavioral feature "total number of steps" and the time discount rate is "0.012" and the significance probability is "0.991." Similarly, the correlation coefficient between the behavioral feature "sleep onset-weight measurement: transition probability" and the time discount rate is "0.450" and the significance probability is "0.012."

[0096] The behavioral feature correlation calculation unit 16 calculates the test statistic for the correlation coefficient obtained in the process (S830), calculates the significance probability, and stores it in the format shown in FIG. 20 (S840).

[0097] The behavioral feature correlation calculation unit 16 passes the behavioral statistical feature data, behavioral transition feature data, behavioral variance feature data, and correlation coefficient and significance probability data of each behavioral feature (see FIG. 20) to the feature selection unit 17 (S850).

[0098] (Detailed processing of feature selection section) The detailed processing of the feature selection unit 17 will be described with reference to Fig. 21. Fig. 21 is a flowchart showing the processing of the feature selection unit.

[0099] First, the feature selection unit 17 receives the behavior statistical feature data, the behavior transition feature data, the behavior variance amount feature data, and the correlation coefficient and significance probability data of each feature from the behavior feature correlation calculation unit 16 (S900).

[0100] The feature selection unit 17 initializes the feature set to be adopted as an empty set (S910).

[0101] The feature selection unit 17 refers to the correlation coefficient and significance probability data of each feature, extracts the behavioral feature with the lowest significance probability, adds it to the set of features to be adopted, and deletes this feature from the correlation coefficient and significance probability data of each feature (S920).

[0102] The feature selection unit 17 deletes (S930) features that have a similar tendency to the features newly added in step S920. Specifically, the feature selection unit 17 calculates the absolute value of the Pearson correlation coefficient between the feature data added in step S920 and each of the feature data, i.e., the behavioral statistical feature data, the behavioral transition feature data, and the behavioral variance amount feature data, and determines whether the absolute value is higher than a threshold set by the system administrator. Features higher than the threshold, i.e., behavioral features determined to have a similar tendency to the features added in step S920, are deleted from the correlation coefficient and significance probability data of each behavioral feature (see FIG. 20). For example, if the selected behavioral feature is "sleep onset-weight measurement: transition probability," "sleep onset-exercise onset: transition probability" is deleted from the correlation coefficient and significance probability data of each behavioral feature in FIG. 20 as a behavioral feature with a similar tendency to this behavioral feature.

[0103] If all data has been deleted from the correlation coefficient and significance probability data, or if it is determined that only features with significance probabilities higher than the threshold value set by the system administrator remain (S940; YES), the feature selection unit 17 proceeds to step S950. Otherwise (S940; NO), the process returns to step S920.

[0104] The feature selection unit 17 extracts only feature data included in the feature set to be adopted from the behavioral statistical feature data, behavioral transition feature data, and behavioral variance feature data, and combines them by user ID (S950). This combined data is treated as various behavioral feature data. In this way, the feature selection unit 17 can select a predetermined number of multiple behavioral features that are correlated with the time discount rate and do not tend to be similar.

[0105] The feature selection unit 17 passes the various behavioral feature data to the model learning and selection unit 18 (S960).

[0106] (Detailed processing of model learning and selection) The detailed processing of the model learning and selection unit 18 will be described with reference to Fig. 22. Fig. 22 is a flowchart showing the processing of the model learning and selection unit.

[0107] First, the model learning and selection unit 18 receives various behavioral feature data from the feature selection unit 17, and receives time discount rate data from the time discount rate data DB 22 (S1000).

[0108] The model learning and selection unit 18 standardizes the feature data (columns) of various behavioral features (SS1010). Specifically, the system administrator selects the model learning and selection unit 18 from the following two methods. Method (1): The model learning and selection unit 18 calculates the mean value μ and standard deviation σ of the data, and converts each value x into a z-score by (x−μ)÷σ. Method (2): The model learning and selection unit 18 finds the maximum value max and minimum value min of the data, and converts each value x by (x-min)÷(max-min).

[0109] This allows values ​​on different scales, such as the behavioral feature "total number of steps" being 6,000 steps and the behavioral feature "sleep time" being 6.2 hours, to be treated as values ​​on the same scale, as they follow a normal distribution with mean μ=0, standard deviation σ=1, and mean 0.

[0110] The model learning and selection unit 18 stores the average value and standard deviation (or maximum and minimum values) obtained from the various behavioral feature data (columns) at this time as model parameters to be used in the estimation phase in the time discount rate estimation model DB 24. The model parameters indicate coefficients that serve as weights by which each standardized value is multiplied after the multiple behavioral features selected by the feature selection unit 17 are standardized.

[0111] Next, the system administrator selects multiple existing supervised machine learning methods, and the model learning and selection unit 18 accepts this selection (S1020). For example, the system administrator may implement a supervised machine learning method or use a library provided in a programming language. Representative examples of supervised machine learning methods include well-known techniques such as linear regression, support vector regression, and decision trees.

[0112] The model learning and selection unit 18 initializes a table in the evaluation value storage DB (see FIG. 6) for each machine learning model prepared by the system administrator. The evaluation value is calculated using known techniques such as "mean absolute error (MAE)" and "mean squared error (MSE)" (S1030).

[0113] The model learning and selection unit 18 combines the various behavioral feature data and the time discount rate data by user ID and divides the data into K sets (S1040). At this time, the system administrator selects one of the following two methods as the data division method: Method [1]: The model learning and selection unit 18 divides the data so that each set has the same number of data items. When a sufficient number of data items can be prepared (for example, several hundred items or more), it is desirable to select this method. Method [2]: The model learning and selection unit 18 divides the data so that each set contains one data item, that is, so that the total number of data items is the same as the number of sets K. When a sufficient number of data items cannot be prepared (for example, less than 100 items), it is desirable to select this method.

[0114] The model learning and selection unit 18 evaluates the accuracy of the (time discount rate estimation) model and stores the evaluation value of each model in the evaluation value storage DB 23 (S1050). Specifically, the model learning and selection unit 18 divides the data set divided in step (S1040) into one set as evaluation data and the remaining sets (K-1) as training data. The model learning and selection unit 18 trains the supervised machine learning model selected in step (S1020) using training data (correct answer data) obtained from survey results, etc., inputs the evaluation data into the obtained model, and calculates the error between the output result and the correct answer data using the evaluation index in step (S1030). The model learning and selection unit 18 repeats this step (S1050) until all sets are used as evaluation data, averages the obtained evaluation indexes, and stores them in the evaluation value storage DB 23 together with each machine learning model.

[0115] The model learning and selection unit 18 refers to the evaluation value storage DB 23 table and selects the model with the best evaluation value (smallest error) (S1060). At this time, if there is a model that has recorded the minimum values ​​for all evaluation indices, it is automatically selected. If the model that obtained the minimum value differs depending on the evaluation indices, the system administrator selects a model according to the following policy: Policy 1: If a system administrator wants to obtain a model with fewer outliers in the prediction results, he or she should select the model that achieves the smallest value in MSE (Mean Squared Error). Policy <2>: In cases other than Policy <1>, the system administrator selects the model that achieves the smallest value in MAE (Mean Absolute Error).

[0116] The model learning / selection unit 18 outputs the supervised learning method and its parameters of the model selected in the process (S1060), as well as the parameters in the process (S1010), to the time discounted estimation model DB 24 (S1070).

[0117] In this way, the time discount factor estimation device 1 can determine model parameters to be assigned to each behavioral feature so as to minimize the error with respect to the time discount factor that is the correct data.

[0118] (Details of the time discount rate estimation section) The detailed processing of time discount rate estimation unit 19 will be described with reference to Fig. 23. Fig. 23 is a flowchart showing the processing of time discount rate estimation unit 19.

[0119] The time discount factor estimation unit 19 receives various behavioral feature data obtained by processing the input data from the behavioral variance calculation unit 15 (S1100).

[0120] The time discount rate estimation unit 19 receives the trained time discount rate estimation model and the parameters of each feature for data standardization (see step S1010) from the time discount rate estimation model DB 24 (S1110).

[0121] The time discount rate estimation unit 19 standardizes various behavioral characteristic data in the same manner as in the processing (S1010). At this time, the time discount rate estimation unit 19 deletes behavioral characteristic data that is not subject to standardization (S1120).

[0122] The time discount rate estimation unit 19 uses the time discount rate estimation model to calculate and output a time discount rate from the various standardized behavioral feature data (S1130).

[0123] [Major Effects of the Embodiments] As described above, according to this embodiment, the time discount factor estimation device 1 can estimate a time discount factor from behavior observed using a wearable device or the like, thereby achieving the effect of being able to estimate an individual's time discount factor with high accuracy without relying on a measurement method using a questionnaire.

[0124] Furthermore, the behavioral data preprocessing unit 11 processes conversion and aggregation of the behavioral data, thereby making it easier for the behavioral statistics calculation unit 12 and the behavioral difference calculation unit 15 to handle the behavioral data. Similarly, the behavioral date and time preprocessing unit 13 processes conversion and aggregation of the behavioral data, thereby making it easier for the behavioral transition calculation unit 14 and the behavioral difference calculation unit 15 to handle the behavioral data.

[0125] Furthermore, the behavioral data preprocessing unit 11 processes behavioral data that can be observed quantitatively statistically (using statistical quantities), and the behavioral transition calculation unit 14 extracts transition features between behaviors that can be observed as event dates and times and uses them for model learning, thereby achieving the effect of being able to estimate the user's time discount rate with high accuracy.

[0126] Furthermore, the behavioral difference calculation unit 15 outputs behavioral difference feature data (see FIG. 18) compiled for each user, which has the effect that the model learning / selection unit 18 can then perform machine learning using a relatively easy known machine learning method such as linear regression, support vector regression, or decision tree.

[0127] Furthermore, the feature selection unit 17 performs a correlation analysis of the time discount rate in advance for the feature data obtained from the behavioral data to extract features with high correlation, and then deletes features with similar tendencies, thereby reducing the dimension of the feature data that serves as input data and making the model lighter.

[0128] In addition, the time discount rate estimation unit 19 outputs as an analysis result which date and time (time) of behavior has a strong influence on the time discount rate estimated from the series of behavioral data, thereby providing the effect of making it possible to provide interpretability for the estimation result.

[0129] 〔supplement〕 The present invention is not limited to the above-described embodiment, and may have the following configurations or processes (operations).

[0130] Each functional configuration of the time discount rate analysis device 1 can be realized by a computer and a program as described above, but this program can also be recorded on a (non-temporary) recording medium and provided, or provided via a network such as the Internet. [Explanation of symbols]

[0131] 1 Time Discount Rate Analyzer 11 Behavioral data preprocessing section 12 Behavioral statistics calculation section 13 Action date and time preprocessing section 14 Behavior transition calculation section 15 Behavior difference calculation unit 16 Behavioral feature correlation calculation unit 17 Feature Selection Section 18 Model learning and selection unit (an example of a model learning unit) 19 Time discount rate estimation part 21 Behavioral Data DB 22 Hourly discount rate data DB 23 Evaluation value storage DB 24-hour discount rate estimation model DB 25 Unknown user behavior data DB

Claims

1. A time discount rate estimation device that estimates a time discount rate in a learning phase, comprising: a model learning unit that calculates an error between a value obtained by standardizing each value related to a plurality of behavioral features of a predetermined user and then multiplying the standardized value by a model parameter indicating each coefficient used as a weighting, and a time discount rate that is correct data based on an answer by the predetermined user, and machine-learns the model parameter so as to reduce the error; a behavioral data preprocessing unit that generates preprocessed behavioral data by calculating summary statistics for each attribute value and the duration of each behavior based on behavior amount data that numerically indicates the behavior of the predetermined user over time and behavior date and time data that indicates the type of behavior of the predetermined user and the date and time of occurrence of the behavior related to that type; a behavioral statistics calculation unit that generates behavioral statistical feature data by calculating an average value for each behavior amount based on the preprocessed behavior data; an action date and time preprocessing unit that generates preprocessed action date and time data based on the action date and time data, the preprocessed action date and time data including an action relationship indicating a relationship between a current action and a next action, and indicating start dates and times of other types of actions that occurred between the current action and the start of the next action of the same type; a behavior transition calculation unit that calculates a transition probability between the current behavior and the next behavior based on the preprocessed behavior date and time data, thereby generating behavior transition feature data; a behavioral difference calculation unit that performs labeling for dividing the preprocessed behavior data at predetermined dates based on date and time information, divides the behavioral statistical feature data based on the labels, and generates difference amount data related to behavior statistics by calculating absolute values ​​of difference amounts of behavior statistics for each divided data, and performs labeling for dividing the preprocessed behavioral date and time data at predetermined dates based on date and time information, divides the behavioral transition feature data based on the labels, and generates difference amount data related to behavior transitions by calculating absolute values ​​of difference amounts of behavior statistics for each divided data, and generates behavioral difference amount feature data by combining the difference amount data related to behavior statistics and the behavioral transition feature data for each user; a behavioral feature correlation calculation unit that calculates a Pearson correlation coefficient between a time discount rate related to the time discount rate data as the correct answer data, and the behavioral statistical feature data, the behavioral transition feature data, and the behavioral variance amount feature data related to the predetermined user, and calculates a test statistic for the correlation coefficient; a feature selection unit that selects, from among the behavioral features related to the behavioral statistical feature data, the behavioral transition feature data, and the behavioral difference amount feature data, a plurality of behavioral features that are correlated with the time discount rate related to the correct answer data and do not have similar tendencies, and combines the selected behavioral features for each user to generate various behavioral feature data; and The model learning unit standardizes the values ​​of the plurality of behavioral features selected by the feature selection unit.

2. The time discount factor estimation device according to claim 1 , wherein the plurality of behavioral characteristics of the predetermined user are based on behavioral data observed by a wearable device attached to the predetermined user.

3. A machine learning method for machine learning model parameters for estimating a time discount rate in a learning phase, comprising: The computer a learning process in which each value relating to a plurality of behavioral features of a predetermined user is standardized, and then each standardized value is multiplied by a model parameter indicating each coefficient used as a weighting, to calculate an error between the value obtained and a time discount rate that is correct data based on an answer by the predetermined user, and the model parameter is machine-learned to reduce the error; behavioral data preprocessing, which generates preprocessed behavioral data by calculating summary statistics for each attribute value and the duration of each behavior based on behavior amount data that numerically indicates the behavior of the predetermined user over time and behavior date and time data that indicates the type of behavior of the predetermined user and the date and time of occurrence of the behavior related to that type; a behavioral statistics calculation process for generating behavioral statistical feature data by calculating an average value for each behavior amount based on the preprocessed behavior data; action date and time preprocessing for generating preprocessed action date and time data based on the action date and time data, the preprocessed action date and time data including an action relationship indicating a relationship between a current action and a next action, and indicating start dates and times of other types of actions that occurred between the current action and the start of the next action of the same type; a behavior transition calculation process for calculating a transition probability between the current behavior and the next behavior based on the preprocessed behavior date and time data, thereby generating behavior transition feature data; a behavioral difference calculation process of labeling the preprocessed behavioral data for division at a predetermined date based on date and time information, dividing the behavioral statistical feature data based on the label, and calculating the absolute value of the difference in the behavioral statistics for each divided data to generate difference amount data related to behavior statistics, and labeling the preprocessed behavioral date and time data for division at a predetermined date based on date and time information, dividing the behavioral transition feature data based on the label, and calculating the absolute value of the difference in the behavioral statistics for each divided data to generate difference amount data related to behavior transitions, and combining the difference amount data related to behavior statistics and the behavioral transition feature data for the same user to generate behavioral difference amount feature data; a behavioral feature correlation calculation process for calculating a Pearson correlation coefficient between a time discount rate related to the time discount rate data as the correct answer data and the behavioral statistical feature data, the behavioral transition feature data, and the behavioral variance amount feature data related to the predetermined user, and calculating a test statistic for the correlation coefficient; a feature selection process for selecting, from among the behavioral features related to the behavioral statistical feature data, the behavioral transition feature data, and the behavioral difference amount feature data, a plurality of behavioral features that are correlated with the time discount rate related to the correct answer data and do not have similar tendencies, and combining the selected behavioral features for each user to generate various behavioral feature data; Run the learning process includes a process of standardizing values ​​of the plurality of behavioral features selected by the feature selection process; Machine learning methods.

4. A program causing a computer to execute the method according to claim 3.

Images