Time discount factor estimation device, machine learning method, time discount factor analysis method, and program

Abstract

The purpose of the present disclosure is to accurately estimate a user's time discount rate without relying on a measurement method that uses a questionnaire.　For this purpose, the disclosed time discount rate estimation device, which estimates a time discount rate in the learning phase, comprises: an action transition time calculation unit that calculates the transition times from a prescribed user's action recorded at each date and time to all types of actions of the prescribed user and thereby outputs action transition time feature data for the action recorded at each date and time; and a time discount rate estimation model training unit that calculates the error between a time discount rate value obtained by entering the action transition time feature data into a deep learning time discount rate estimation model and a time discount rate that is correct answer data based on the prescribed user's answer, and trains the time discount rate estimation model through machine learning so that the error is reduced.

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 an individual'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 time discount rate estimation device having: a behavior transition time calculation unit that calculates transition times to all types of behavior of a specified user for behavior recorded at each date and time of the specified user, and outputs behavior transition time feature data for each behavior recorded at the said date and time; and a time discount rate estimation model learning unit that calculates the error between a time discount rate value obtained by inputting the behavior transition time feature data into a time discount rate estimation model using deep learning and a time discount rate that is correct data based on answers by the specified user, and machine-learns the time discount rate estimation model so as to reduce the error. [Effects of the Invention]

[0009] As described above, the present invention has the effect of enabling an individual'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. 1 is a hardware configuration diagram of a time discount factor estimation device according to an embodiment. [Figure 4] FIG. 2 is a conceptual diagram of tables constituting a behavior data DB. [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 tables constituting a time discount rate estimation model 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 an outline of a process for estimating a time discount rate in the estimation phase. [Figure 10] FIG. 10 is a conceptual diagram illustrating an example of output from the behavioral data preprocessing unit. [Figure 11] 10 is a flowchart showing the processing of a behavior transition time calculation unit. [Figure 12] FIG. 10 is a conceptual diagram showing an example of output from a behavior transition time calculation unit. [Figure 13] FIG. 10 is a diagram showing a network structure of a time discount rate estimation model constructed by a time discount rate estimation model construction unit. [Figure 14] FIG. 1 is a diagram illustrating a computational image of a self-attention mechanism. [Figure 15] 10 is a flowchart showing the processing of a time discount factor estimation model learning unit. [Figure 16] 10 is a flowchart showing the processing of a time discount factor estimation unit. [Figure 17] 10 is a flowchart showing the processing of an estimation result interpretation unit. [Figure 18] FIG. 10 is a diagram illustrating an example of a visualization output from an estimation result interpretation 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 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 such as a questionnaire.

[0013] [Hardware Configuration of the Time Discount Factor Estimation Device] Next, the hardware configuration of the time discount factor estimation device 1 will be described with reference to Fig. 3. Fig. 3 is a diagram showing the hardware configuration of the time discount factor estimation device according to the embodiment.

[0014] 3, time discount factor estimation device 1 includes processor 101, memory 102, auxiliary storage device 103, connection device 104, communication device 105, and drive device 106. The hardware components constituting time discount factor estimation device 1 are interconnected via bus 107.

[0015] Processor 101 serves as a control unit that controls the entire time discount factor estimation device 1, and includes various arithmetic 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 estimation 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 estimation 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 transition time calculation unit 12, a time discount factor estimation model construction unit 17, and a time discount factor estimation model learning 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, 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 transition time calculation unit 12, a time discount factor estimation unit 19, and an estimation result interpretation unit 20. Each of these units is a function realized by instructions from a processor 101 (shown in FIG. 3, which will be described later) based on a program.

[0026] Furthermore, in the estimation phase, time discount factor estimation device 1 has a time discount factor estimation model DB 24. This time discount factor estimation model DB is constructed in memory 102 or auxiliary storage device 203. In the learning phase, time discount factor estimation device 1 uses information from each DB to output a learned time discount factor estimation model.

[0027] <Behavioral Data DB> FIG. 4 is a conceptual diagram showing tables that make up the behavioral data DB. The behavioral data BD21 stores, for a user ID, behaviors that are automatically recorded by a wearable device or recorded by the user themselves, along with a character string that represents the date and time of the user's behavior identified by the user ID and the type (content) of the behavior. The type of behavior may be stored in the behavioral 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] <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.

[0029] <Time discount rate estimation model DB> 6 is a conceptual diagram of tables constituting the time discount factor estimation model DB. In the time discount factor estimation model DB 24, parameter values ​​associated with each parameter name are managed for each machine learning model meter name.

[0030] <Each function configuration> Next, each functional configuration of the time discount factor estimation device 1 in the learning phase will be described.

[0031] The behavioral data preprocessing unit 11 deletes data relating to the same type of behavior observed continuously within a predetermined time period from the behavioral data, and then assigns a unique behavior ID corresponding to the type of behavior, and associates this behavior ID with the behavior transition time feature data, thereby preprocessing the behavioral data.

[0032] The behavior transition time calculation unit 12 calculates the transition time to all types of behavior of a specified user for the behavior recorded at each date and time of the specified user, and outputs behavior transition time characteristic data for each behavior recorded at each date and time.

[0033] The time discount rate estimation model constructing unit 17 constructs the structure of a time discount rate estimation model as shown in FIG. 13, which will be described later.

[0034] The time discount rate estimation model learning unit 18 calculates the error between the value of the time discount rate obtained by inputting behavioral transition time feature data into a time discount rate estimation model using a DNN (Deep Neural Network: deep learning) and the time discount rate that is correct data based on answers from a specified user, and performs machine learning on the time discount rate estimation model so as to reduce this error.

[0035] The time discount rate estimation unit 19 uses machine-learned model parameters (time discount rate estimation model) to calculate and output a time discount rate based on behavioral data (input data) indicating the behavior of a specific user recorded at each date and time.

[0036] The estimation result interpretation unit 20 visualizes and outputs the importance of the behavior recorded at each date and time of a specific user based on the weight (importance) for each transition time. Each of the above components will be described in detail later.

[0037] [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.

[0038] <Processing Overview> (Outline of the learning phase process) FIG. 7 is a flowchart showing an outline of the process of estimating the time discount factor in the learning phase.

[0039] First, the behavioral data preprocessing unit 11 receives and processes the behavioral data of each person (see FIG. 4) from the behavioral data DB 21 (S100). The details of this processing will be described later.

[0040] The behavior transition time calculation unit 12 receives and processes the preprocessed behavior data from the behavior data preprocessing unit 11 (S110). Details of the processing will be described later. FIG. 12 shows an example of data obtained as the output of the behavior transition time calculation unit 12. As shown in FIG. 12, the output data of the behavior transition time calculation unit 12 associates a user ID, the date and time of occurrence of the behavior, the behavior content (type), the behavior ID, and behavior transition time feature data. Each transition time indicates the difference in time between the start date and time of the behavior and the start date and time of another behavior. Note that, as can be seen from the fact that multiple user IDs "001" are managed here, multiple behavior transition times for one user are shown.

[0041] The time discount rate estimation model constructing unit 17 constructs a time discount rate estimation model (S120). Details of this process will be described later.

[0042] The time discount rate estimation model learning unit 18 receives behavioral transition time feature data from the behavioral transition time calculation unit 12, receives time discount rate data as correct answer data for machine learning from the time discount rate data DB22, receives a time discount rate estimation model from the time discount rate estimation model construction unit 17, learns the model, and outputs the learned model to the time discount rate estimation model DB24.

[0043] (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.

[0044] First, the behavioral data preprocessing unit 11 receives and processes a user's behavioral data sequence as input (S200).

[0045] The behavior transition time calculation unit 12 receives and processes the preprocessed behavior data from the behavior data preprocessing unit 11 (S210).

[0046] The time discount factor estimation unit 19 receives the trained model from the time discount factor estimation model DB 24, calculates the time discount factor, and outputs it (S220). Details of this process will be described later.

[0047] The estimation result interpretation unit 20 receives and processes the parameter set obtained during estimation from the time discount factor estimation unit 19, and outputs the analysis result (S230). Details of the processing will be described later.

[0048] <Detailed processing> Next, the detailed processing of the learning phase will be described.

[0049] (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.

[0050] First, the behavioral data preprocessing unit 11 receives a behavioral data series as shown in FIG. 4 as an example of user behavioral data from the behavioral data DB 21 in the learning phase, or as an input in the estimation phase (S300).

[0051] In Figure 4, the behavioral data preprocessing unit 11 simultaneously scans the "user ID," "date and time," and "behavior" columns, and deletes data related to the same type of behavior observed consecutively within a predetermined short period of time (e.g., 10 minutes). For example, if a user's behavior "starts exercising" is observed multiple times in succession within a short period of time, the behavioral data preprocessing unit 11 will retain only the first observed "starts exercising" behavior and delete the other same behaviors as they were mistakenly observed. The time interval can be set by the system administrator.

[0052] The behavioral data preprocessing unit 11 scans the "behavior" column in Figure 4 and deletes behaviors with a low number of observations. Specifically, the behavioral data preprocessing unit 11 counts the number of occurrences for each type of behavior and deletes behaviors with a number of occurrences below a threshold set by the system administrator. The threshold for the number of occurrences can be set by the system administrator.

[0053] 4, in the learning phase, the behavior data preprocessing unit 11 scans the "behavior" column, memorizes the types of behavior of all users, and assigns a behavior ID indicating a unique numerical value corresponding to the type of behavior (S330). In the estimation phase, this process (S330) is omitted.

[0054] The behavior data preprocessing unit 11 adds an "behavior ID" column and stores numerical values ​​associated with the data in the "behavior" column (see FIG. 10) (S340).

[0055] The behavior data preprocessing unit 11 passes the preprocessed behavior data (see FIG. 10) converted by the process (S340) to the behavior transition time calculation unit 12 (S350).

[0056] (Detailed processing of the behavior transition time calculation part) Detailed processing of the behavior transition time calculation unit 12 will be described with reference to Fig. 11. Fig. 11 is a flowchart showing processing of the behavior transition time calculation unit.

[0057] First, the behavior transition time calculation unit 12 receives the converted preprocessed behavior data from the behavior data preprocessing unit 11 (S400).

[0058] The behavior transition time calculation unit 12 aggregates data for each "user ID" in FIG. 10, and further calculates the average value for each behavior amount (each column) (S410).

[0059] The behavior transition time calculation unit 12 aggregates data for each "user ID," calculates the transition time to all types of behavior for the behavior recorded at each date and time, and stores the transition time for each type of behavior in the memory 102 (S420). Specifically, when targeting behavior at a certain date and time, the behavior transition time calculation unit 12 scans the data after that date and time, extracts the dates and times when all types of behavior are first observed, and calculates the difference between them. For behaviors that are not observed after that date and time, the behavior transition time calculation unit 12 stores a value indicating a missing value, such as NULL, in the memory 102.

[0060] The behavioral transition time calculation unit 12 treats the data obtained in the process (S420) as behavioral transition time feature data (see FIG. 12), and passes it to the time discount rate estimation model learning unit 18 in the learning phase, or to the time discount rate estimation unit 19 in the estimation phase (S430).

[0061] Next, an example of a time discount rate estimation model constructed by the time discount rate estimation model construction unit 17 will be shown using Fig. 13 and Fig. 14. The time discount rate estimation model is constructed using a DNN structure. Fig. 13 is a diagram showing an example of a time discount rate estimation model constructed by the time discount rate estimation model construction unit. Fig. 14 is a diagram showing a calculation image of the self-attention mechanism 50.

[0062] The time discount rate estimation model receives behavioral transition time feature data of a specific user as input data and generates time discount rate data of the same specific user as output data. The DNN network structure of the time discount rate estimation model consists of the following units.

[0063] The first is the embedding layer 31, which extracts abstract features from the behavior ID. The embedding layer 31 converts the behavior ID in Fig. 12 into a one-hot representation with the same number of dimensions as the number of types of behavior, and then converts it into a feature vector with dimensions determined by the system administrator.

[0064] The second is a first fully connected layer 32 that extracts abstract features from the behavior transition time feature data in FIG. 12. The first fully connected layer 32 performs nonlinear transformation on the feature quantities of the input data using, for example, a sigmoid function or a ReLu function, to obtain a feature vector. Note that the first input in FIG. 13 is the "behavior ID" and "behavior transition time feature data" of the top record in FIG. 12. Furthermore, the second input in FIG. 13 is the "behavior ID" and "behavior transition time feature data" of the second record from the top in FIG. 12. In this manner, input is continued up to the final input for the same user ID.

[0065] The third is a long-short term memory (LSTM), which further abstracts the abstracted 64-dimensional feature vector as sequence data. Specifically, each of the multiple LSTMs, 40-1, 40-2, ..., 40-T, sequentially receives sequence data at each time step and repeatedly performs nonlinear transformations while taking into account past abstracted information. Note that any of the multiple LSTMs, 40-1, 40-2, ..., 40-T, will be referred to as LSTM40.

[0066] The fourth is a self-attention mechanism (Self-Attention) 50 that calculates a weighted average to obtain a feature vector that takes into account the importance of the set of abstracted feature vectors by LSTM 40. The weighting calculation is achieved by two fully connected layers. Here, the first layer, the second fully connected layer 60a, takes each feature vector abstracted by LSTM as input and outputs a context vector of any size. The second layer, the second fully connected layer 60b, takes the context vector as input and outputs a scalar value corresponding to the importance. The context vector may be subjected to a nonlinear transformation. The importance is converted to a value corresponding to a probability value using, for example, a softmax function.

[0067] The fifth is a second fully connected layer 60 that converts the feature vector weighted and averaged by the self-attention mechanism 50 into a scalar value corresponding to the temporal discount rate.

[0068] Here, we will explain the calculation image of the self-attention mechanism 50 using Figure 14. In Figure 14, the 64-dimensional output vector is simplified and shown as a 4-dimensional output vector. Also, the size of the output vector of each LSTM can be adjusted arbitrarily.

[0069] As shown in Figure 14, the self-attention mechanism 50 calculates the weights for each time step based on the output vectors of the LSTM 40 for each time step (1), (2), ... (T) (S1). Here, the weight for time step (1) is shown as "0.0001." These weights are also used by the inference result interpretation unit 20.

[0070] Next, the self-attention mechanism 50 calculates the weighted average (S2). For example, at time step (1), the weight 0.0001 × output vector {0.1, 0.2, 0.5, 10.2} = {0.00001, 0.00002, 0.00005, 0.00102}, and at time step (2), the weight 0.02 × output vector {0.4, 0.5, 1.5, 0.1} = {0.008, 0.01, 0.03, 0.00}. This calculation continues until time step (T). Then, the self-attention mechanism 50 adds the vector values ​​for each dimension to obtain output data with the same number of dimensions as the output vector of the LSTM 40. For example, when adding all the values ​​in one dimension, as shown in Figure 14, the result is 0.4 + 0.008 + ··· = 0.84. Similarly, adding up all two-dimensional values ​​gives 0.09, adding up all three-dimensional values ​​gives 0.20, and adding up all four-dimensional values ​​gives 0.10. In this way, the behavior transition time feature data based on the 64-dimensional feature vector includes a behavior ID indicating a behavior such as starting to sleep, and behavior transition time feature data that is the transition time to each of the other behaviors, as shown in Fig. 12 .

[0071] (Details of the time discount rate estimation model learning section) Detailed processing of the time discount rate estimation model learning unit 18 will be described with reference to Fig. 15. Fig. 15 is a flowchart showing processing of the time discount rate estimation model learning unit.

[0072] As shown in FIG. 15, the time discount rate estimation model learning unit 18 receives behavioral transition time feature data from the behavioral transition time calculation unit 12, receives time discount rate data as correct answer data from the time discount rate data DB 22, and associates the data by user ID (S500).

[0073] The time discount factor estimation model learning unit 18 receives the DNN network structure (framework) as shown in FIG. 13 from the time discount factor estimation model construction unit 17 (S510).

[0074] The time discount factor estimation model learning unit 18 initializes the model parameters of each unit in the network structure (S520). For example, the time discount factor estimation model learning unit 18 initializes the parameters with random numbers between 0 and 1.

[0075] The time discount rate estimation model learning unit 18 learns and updates the time discount rate estimation model (model parameters) using the time discount rate data corresponding to the behavior transition time feature data for each user ID (S530). The parameter learning is performed by machine learning the time discount rate estimation model (model parameters) using a known technique such as backpropagation so as to reduce the error between the time discount rate value output by the second fully connected layer 60 and the time discount rate data as correct data.

[0076] Time discount rate estimation model learning unit 18 outputs the learned time discount rate estimation model (network structure (see FIG. 13) and model parameters (see FIG. 6)), and stores the output result in time discount rate estimation model DB 24.

[0077] (Details of the time discount rate estimation section) Detailed processing of the time discount rate estimation unit 19 will be described with reference to Fig. 16. Fig. 16 is a flowchart showing processing of the time discount rate estimation unit.

[0078] First, the time discount rate estimation unit 19 receives, from the behavior transition time calculation unit 12, behavior transition time feature data that the behavior transition time calculation unit 12 has obtained by processing input data (S600).

[0079] The time discount rate estimation unit 19 receives the trained time discount rate estimation model from the time discount rate estimation model DB 24 (S610).

[0080] The time discount rate estimation unit 19 calculates and outputs the time discount rate from the behavior transition time feature data using the trained time discount rate estimation model (S620).

[0081] The time discount factor estimation unit 19 transfers the importance of the self-attention mechanism in the trained time discount factor estimation model obtained for the input data to the estimation result interpretation unit 20 in association with the input data (S630).

[0082] (Detailed processing of the estimation result interpretation section) Detailed processing of the estimation result interpretation unit 20 will be described with reference to Fig. 17. Fig. 17 is a flowchart showing processing of the estimation result interpretation unit.

[0083] First, the estimation result interpretation unit 20 receives the importance of the self-attention mechanism associated with the input data ("weight" in FIG. 14) from the time discount factor estimation unit 19 (S700).

[0084] The estimation result interpretation unit 20 visualizes and outputs the estimated importance together with date and time information and each behavior (S710). FIG. 18 is a diagram showing an example of the visualization output by the estimation result interpretation unit 20. In FIG. 18, the horizontal axis represents date and time information and the behavior (type) at that time, and the vertical axis represents the importance value, which is visualized as a line graph with respect to the date and time information. That is, the graph in FIG. 18 visualizes the degree to which behavior at each date and time contributes to the time discount rate. This visualization is possible because, although a complex network structure like that in FIG. 13 is used, each time step (behavior at each date and time) is not aggregated for each user as in FIGS. 13 and 14, but each time step is input directly into the time discount rate estimation model. For example, a user with a high time discount rate tends to think negatively. However, FIG. 18 shows that the importance (contribution) of starting sleep is high. Therefore, the user can try to lower their time discount rate by starting sleep earlier or later. As a result, if the time discount rate output by the time discount rate estimation unit 19 were the only information available to the user, the user would only be able to grasp whether he or she tends to think positively or negatively. In contrast, the analysis results output by the estimation result interpretation unit 20 allow the user to grasp what behavior in his or her lifestyle habits he or she needs to change in order to be able to think positively.

[0085] [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.

[0086] Furthermore, the behavior data preprocessing unit 11 processes the behavior data, such as conversion and aggregation, to make it easier for the behavior transition time calculation unit 12 to handle the behavior data.

[0087] Furthermore, the time discount rate estimation model learning unit 18 processes the behavioral data as sequential data using a DNN time discount rate estimation model such as that shown in FIG. 13, which enables the time discount rate estimation unit 19 to extract features that take into account the context of the behavior, thereby achieving the effect of being able to estimate the user's time discount rate with high accuracy.

[0088] Furthermore, the behavioral transition time calculation unit 12 calculates the transition time between each behavior for the behavioral data and uses the calculated transition time as an input feature, which enables the time discount rate estimation unit 19 to take into account the transition relationship between the behaviors, thereby achieving the effect of being able to estimate the user's time discount rate with high accuracy.

[0089] In addition, the time discount rate estimation unit 19 outputs, as an analysis result, which date and time actions have a strong influence on the time discount rate estimated from the series of behavioral data, thereby providing the effect of making the estimation result interpretable.

[0090] 〔supplement〕 The present invention is not limited to the above-described embodiment, and may have the following configurations or processes (operations).

[0091] As described above, each functional configuration of the time discount factor estimation device 1 can be realized by a computer and a program, but this program can also be recorded on a (non-temporary) recording medium and provided, or can be provided via a network such as the Internet. [Explanation of symbols]

[0092] 1 Hourly discount rate estimator 11 Behavioral data preprocessing section 12 Behavior transition time calculation section 17 Time discount rate estimation model construction section 18 Hourly discount rate estimation model training part 19 Time discount rate estimation part 20 Estimation result interpretation section 21 Behavioral Data DB 22 Hourly discount rate data DB 24-hour discount rate estimation model DB

Claims

1. A time discount rate estimation device that estimates a time discount rate in a learning phase, comprising: a behavior transition time calculation unit that calculates transition times to all types of behavior of a predetermined user for the behavior recorded at each date and time of the predetermined user, and outputs behavior transition time feature data for each behavior recorded at the respective date and time; a time discount rate estimation model learning unit that calculates an error between a time discount rate value obtained by inputting the behavioral transition time feature data into a time discount rate estimation model based on deep learning and a time discount rate that is correct answer data based on an answer by the predetermined user, and machine-learns the time discount rate estimation model so as to reduce the error; A time discount rate estimation device having the following.

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

3. 3. The time discount rate estimation device according to claim 2, a behavioral data pre-processing unit that pre-processes the behavioral data by deleting data relating to the same type of behavior observed continuously over a predetermined period of time from the behavioral data, and then assigning unique behavior identification information corresponding to the type of behavior, and associating the behavior identification information with the behavior transition time feature data.

4. A time discount rate estimation device that estimates a time discount rate in an estimation phase, comprising: A time discount rate estimation device having a time discount rate estimation unit that calculates the error between the value of a time discount rate obtained by inputting behavioral transition time feature data indicating the transition time to all types of behavior of a specified user for behavior recorded at each date and time into a time discount rate estimation model using deep learning, and the time discount rate that is correct data based on the answers of the specified user, and calculates and outputs a time discount rate based on behavioral data indicating the behavior of a specific user recorded at each date and time using a machine-learned time discount rate estimation model obtained by machine learning so as to reduce the error.

5. 5. The time discount rate estimation device according to claim 4, the time discount rate estimation model has a self-attention mechanism that calculates a weight for each transition time, A time discount rate estimation device having an estimation result interpretation unit that visualizes and outputs the importance of the behavior recorded at each date and time of the specific user based on the weight for each transition time.

6. A machine learning method for machine learning a time discount rate estimation model for estimating a time discount rate in a learning phase, comprising: The computer Calculating transition times to all types of behaviors of a specific user for the behaviors recorded at each date and time of the specific user, and outputting behavior transition time feature data for each behavior recorded at the date and time; The time discount rate estimation model based on deep learning is trained by machine learning to calculate the error between the value of the time discount rate obtained by inputting the behavioral transition time feature data and the time discount rate that is correct data based on the answer by the predetermined user, and the time discount rate estimation model is trained by machine learning to reduce the error. Machine learning methods.

7. A time discount rate estimation method for estimating a time discount rate in an estimation phase, comprising: The computer A time discount rate estimation method that calculates the error between the time discount rate value obtained by inputting behavioral transition time feature data showing the transition time to all types of behavior of a specified user for the behavior recorded at each date and time into a time discount rate estimation model using deep learning, and the time discount rate that is correct data based on the answers of the specified user, and uses a machine-learned time discount rate estimation model obtained by machine learning to reduce the error, to calculate and output a time discount rate based on behavioral data showing the behavior of a specific user recorded at each date and time.

8. A program that causes a computer to execute the method according to claim 6 or 7.

Images