Prediction device, prediction method, and prediction program

Abstract

This prediction device obtains the result of predicting a psychological index of a user by inputting attribute information and psychological information related to the user to a prediction model trained to receive input of attribute information and psychological information related to a user to output the result of predicting the psychological index of the user. The prediction device performs correction for excluding, from the obtained prediction result, a component linearly correlated with a feature amount of the psychological information related to the user. Specifically, the prediction device corrects the obtained prediction result so that a vector representing the prediction result is orthogonal to a vector representing the feature amount of the psychological information related to the user.

Description

Prediction device, prediction method, and prediction program

[0001] The present invention relates to a prediction device, a prediction method, and a prediction program for predicting a psychological index of a user.

[0002] The scope of application of machine learning is expanding beyond attribute information (objective information about users) to psychological information (subjective information about users obtained through questionnaires, for example).

[0003] For example, consider the case of predicting a person's psychological indicators, such as whether or not a person will resign. In this case, whether or not to resign depends on the person's psychological state, so it is expected that prediction accuracy will be improved by utilizing psychological information rather than the person's attribute information.

[0004] Tournament overview, [online], [Retrieved December 15, 2023], Internet <URL: https: / / jp.docs.numer.ai / numerai-tnamento / tournament-overview>

[0005] Here, psychological information is influenced by mood at the time, which may change the results. Therefore, psychological information is considered to be information with more noise than attribute information. When such psychological information is used as a feature for machine learning, there is a problem that prediction accuracy may decrease. Therefore, an object of the present invention is to perform robust prediction even when psychological information is used as a feature.

[0006] In order to solve the above-mentioned problems, the present invention is characterized by comprising an input unit that accepts input of attribute information indicating a user's attributes and psychological information indicating the user's psychological state, a prediction unit that predicts the user's psychological indicators using a prediction model that is trained to input the user's attribute information and psychological information and output a prediction result of the user's psychological indicators, and a correction unit that corrects the prediction result to exclude components that are linearly correlated with elements that constitute the feature quantities of the user's psychological information and outputs the correction result.

[0007] According to the present invention, prediction can be performed robustly even when psychological information is used as a feature.

[0008] FIG. 1 is a diagram for explaining an overview of a prediction device. FIG. 2 is a diagram for explaining correction of a prediction result y by the prediction device. FIG. 3 is a diagram showing an example of the configuration of a prediction device. FIG. 4 is a flowchart showing an example of a processing procedure executed by the prediction device. FIG. 5 is a diagram showing the results of an experiment for confirming the accuracy of prediction by the prediction device. FIG. 6 is a diagram showing a computer that executes a prediction program.

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a description will be given of an embodiment of the present invention with reference to the drawings, but the present invention is not limited to the embodiment.

[0010] [Overview] First, an overview of the prediction device of this embodiment will be described using Figure 1. In conventional predictions using machine learning models, a user's indicators (e.g., turnover rate, bankruptcy rate, desired items) have been predicted based on the user's attribute information (e.g., age, gender, occupation, years of service, etc.) ((1)). However, if the indicator to be predicted is psychological, it is better to also use the user's psychological information as input to the machine learning model (prediction model). Therefore, the prediction device of this embodiment uses the user's psychological information (e.g., well-being, satisfaction with the company, etc.) in addition to the user's attribute information as input to the prediction model ((2)).

[0011] For example, the prediction device learns a prediction model so that it receives user attribute information and psychological information and outputs a prediction result of the user's psychological index. The prediction result (y) output by this prediction model can be expressed as the following equation (1).

[0012]

[0013] The feature used by the prediction model for prediction is X(x1, x2, ..., x d )∈R n×d where n is the number of data and d is the dimension.

[0014] A trained prediction model may contain components that depend on one feature (x1), such as f(x1) in the above formula (1). When calculating prediction results, if the prediction results depend too much on one feature (x1), the model will be vulnerable when that feature changes. Therefore, when calculating prediction results, it is important not to depend on one feature (x1) but to use as many features (x1, x2, ..., x d ) is a better choice.

[0015] Therefore, when predicting a user's psychological indicators using a prediction model, the prediction device of this embodiment makes the prediction using g, rather than relying solely on f in the above equation (1).

[0016] For example, the prediction device assumes that f in equation (1) is linearly correlated with x and removes the linear component from y. For example, the prediction device calculates w^ based on the following equation (2) so that the difference between y (prediction result) and the value obtained by multiplying X (feature) by a weight w is minimized.

[0017]

[0018] The prediction device then substitutes the calculated w^ into the following equation (3) to remove the linear component from the prediction result y. Note that λ in equation (3) is a hyperparameter that takes a value between 0 and 1.

[0019]

[0020] The process of removing the linear component from the prediction result y as described above can be interpreted as correcting the prediction result y so that it is geometrically orthogonal to the vector representing the feature quantity X (see FIG. 2).

[0021] By correcting the prediction result y as described above, the prediction device can robustly predict psychological indicators even when the user's psychological information, which is thought to be relatively noisy, is used as a feature.

[0022] [Configuration Example] Next, a configuration example of the prediction device 10 will be described with reference to Fig. 3. The prediction device 10 includes, for example, an input / output unit 11, a storage unit 12, and a control unit 13.

[0023] The input / output unit 11 is an interface that controls the input and output of various data. The input / output unit 11 receives input of, for example, user attribute information, psychological information, etc. The input / output unit 11 also outputs the prediction results of the user's psychological indicators.

[0024] The storage unit 12 stores data, programs, etc. that are referenced when the control unit 13 executes various processes. The storage unit 12 is realized by a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. For example, the storage unit 12 stores parameters of a prediction model after learning, etc.

[0025] The prediction model is a machine learning model that receives attribute information and psychological information of a user as input and outputs a prediction result of the psychological index of the user. This prediction model is trained by, for example, a learning unit 131 (described later) of the control unit 13.

[0026] The control unit 13 controls the entire prediction device 10. The functions of the control unit 13 are realized, for example, by a CPU (Central Processing Unit) executing a program stored in the storage unit 12.

[0027] The control unit 13 includes, for example, a learning unit 131, an input unit 132, a prediction unit 133, and a correction unit 134. The learning unit 131 learns the prediction model using training data for the prediction model (for example, a dataset of user attribute information and psychological information and values ​​of the user's psychological indicators). The input unit 132 accepts input of user attribute information, psychological information, etc. to be input to the prediction model.

[0028] The prediction unit 133 inputs the user's attribute information and psychological information into the trained prediction model, thereby obtaining a prediction result of the user's psychological index.

[0029] The correction unit 134 corrects the prediction result of the user's psychological index obtained by the prediction unit 133. For example, the correction unit 134 excludes, from the prediction result of the user's psychological index obtained by the prediction unit 133, a component that is linearly correlated with an element that constitutes the feature amount of the user's psychological information.

[0030] For example, based on the above-mentioned formula (2), the correction unit 134 calculates w^ such that the value obtained by subtracting the value obtained by multiplying X (feature) input to the prediction model by a certain weight w is minimized from y (prediction result) output by the prediction model.

[0031] Then, the correction unit 134 uses the calculated w^ to remove the linear component from the prediction result y based on the above-mentioned equation (3). In other words, the correction unit 134 corrects the prediction result output by the prediction model so that the vector indicating the prediction result is orthogonal to the vector indicating the feature amount of the user's psychological information. Then, the correction unit 134 outputs the corrected prediction result.

[0032] By correcting the prediction result as described above, the prediction device 10 can robustly predict psychological indicators even when the user's psychological information, which is thought to be relatively noisy, is used as a feature.

[0033] [Example of Processing Procedure] Next, an example of processing procedure executed by the prediction device 10 will be described with reference to Fig. 4. First, the learning unit 131 of the prediction device 10 uses a data set of attribute information and psychological information of a user and values ​​of psychological indices of the user as training data to learn a prediction model (S1).

[0034] After that, when the input unit 132 receives input of the user's attribute information and psychological information (S2), the prediction unit 133 inputs the user's attribute information and psychological information into the prediction model trained in S1 and predicts the user's psychological index (S3). Then, the correction unit 134 corrects the prediction result obtained in S3 using the above-mentioned formulas (2) and (3) (S4), and outputs the corrected prediction result (S5).

[0035] [Example of Embodiment] The following describes an example of prediction of a user's psychological indicators performed by the prediction device 10. Here, the prediction device 10 uses the user's everyday conversation as psychological information of the user and predicts the probability that the user will quit their job.

[0036] In this case, the user of the prediction device 10 collects user attribute information and everyday conversations in advance. The user also conducts a survey on the indicator to be predicted (probability of quitting one's job) to collect training data. The learning unit 131 of the prediction device 10 then uses the collected data as training data to learn a prediction model that predicts the probability of a user quitting their job from the user's attribute information and everyday conversations.

[0037] The prediction device 10 then predicts the probability of a user quitting their job by inputting the user's attribute information and psychological information (user conversation) into the trained prediction model. The prediction device 10 then corrects the prediction result (the user's probability of quitting their job) obtained by the prediction model so that it is orthogonal to the vector of the feature quantity of the psychological information (user conversation). The prediction device 10 can robustly predict the probability of a user quitting their job when using psychological information such as the user's everyday conversation.

[0038] [Experimental Results] The following experiment was conducted to confirm the effect of correcting the prediction result (hereinafter referred to as "neutralization") by the prediction device 10. The results of the experiment are shown in FIG.

[0039] In the experiment, we compared the prediction accuracy for three cases: when no neutralization was performed on any of the features input to the prediction model ("no neutralization"), when neutralization was performed on all features ("neutralization on all inputs"), and when neutralization was performed on psychological information only ("neutralization limited to psychological data only"). The experimental settings are as follows:

[0040] Task: Predict meaningful indicators (whether or not someone wants to quit their job, etc.) from the survey information. Because the actual content of the indicators is sensitive, we will simply explain here that we predicted six independent indicators. These indicators are expressed as integers between 1 and 5, so this can be thought of as a five-class classification problem.

[0041] ・Data (training data and test data): Attribute data and psychological data collected by conducting a survey of approximately 70,000 people at a certain company were used. The collected data also included responses to six indicators that were prediction targets. ・Validation: stratified 5-hold ・Handling of categorical variables: label encoding ・Neutralization: performed on 500 pieces of data

[0042] The values ​​shown in Figure 5 are the prediction accuracy and average prediction accuracy for each of the six indices obtained through the experiment. Note that λ (see Equation (3)) used to correct the prediction results was selected to be an appropriate value between 0 and 1. As shown in Figure 5, it was confirmed that prediction accuracy can sometimes be improved with neutralization compared to without neutralization.

[0043] Furthermore, when neutralization was performed by focusing only on psychological data, it was confirmed that the prediction accuracy improved in all cases compared to when neutralization was not performed. This is thought to be because psychological data is prone to contain noise, and the benefits of not placing too much faith in the psychological data outweigh the disadvantage of information degradation due to the elimination of linear correlation.

[0044] [System Configuration, etc.] The components of each unit shown in the figure are conceptual functional units and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc. Furthermore, all or any part of the processing functions performed by each device can be realized by a CPU and a program executed by the CPU, or can be realized as hardware using wired logic.

[0045] Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method.In addition, the information including the processing procedures, control procedures, specific names, various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified.

[0046] [Program] The prediction device 10 can be implemented by installing a program (prediction program) as package software or online software on a desired computer. For example, by executing the program on an information processing device, the information processing device can function as the prediction device 10. The information processing device referred to here includes mobile communication terminals such as smartphones, mobile phones, and PHS (Personal Handyphone Systems), as well as terminals such as PDAs (Personal Digital Assistants).

[0047] 6 is a diagram showing an example of a computer that executes a prediction program. The computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 also includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0048] The memory 1010 includes a read-only memory (ROM) 1011 and a random access memory (RAM) 1012. The ROM 1011 stores a boot program such as a basic input / output system (BIOS). The hard disk drive interface 1030 is connected to a hard disk drive 1090. The disk drive interface 1040 is connected to a disk drive 1100. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to a mouse 1110 and a keyboard 1120, for example. The video adapter 1060 is connected to a display 1130, for example.

[0049] The hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, the programs that define the processes executed by the prediction device 10 are implemented as program modules 1093 in which computer-executable code is written. The program modules 1093 are stored, for example, in the hard disk drive 1090. For example, the program modules 1093 for executing processes similar to those of the functional configuration of the prediction device 10 are stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD (Solid State Drive).

[0050] Data used in the processing of the above-described embodiment is stored as program data 1094, for example, in the memory 1010 or the hard disk drive 1090. The CPU 1020 then reads the program module 1093 or the program data 1094 stored in the memory 1010 or the hard disk drive 1090 into the RAM 1012 as necessary and executes them.

[0051] The program module 1093 and program data 1094 may not necessarily be stored in the hard disk drive 1090, but may also be stored in a removable storage medium and read by the CPU 1020 via the disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (such as a local area network (LAN) or a wide area network (WAN)). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via the network interface 1070.

[0052] REFERENCE SIGNS LIST 10 Prediction device 11 Input / output unit 12 Storage unit 13 Control unit 131 Learning unit 132 Input unit 133 Prediction unit 134 Correction unit

Claims

1. A prediction device comprising: an input unit that receives an input of attribute information indicating a user's attributes and psychological information indicating the user's psychological state; a prediction unit that predicts the user's psychological index using a prediction model learned to output a prediction result of the user's psychological index with the user's attribute information and psychological information as inputs; and a correction unit that performs and outputs a correction for excluding a component linearly correlated with an element constituting a feature amount of the user's psychological information from the prediction result.

2. The prediction device according to claim 1, wherein the correction for the prediction result is a correction performed so that a vector representing the prediction result is orthogonal to a vector representing a feature amount of the user's psychological information.

3. A prediction method executed by a prediction device, the method comprising: a step of receiving an input of attribute information indicating a user's attributes and psychological information indicating the user's psychological state; a step of predicting the user's psychological index using a prediction model learned to output a prediction result of the user's psychological index with the user's attribute information and psychological information as inputs; and a step of performing and outputting a correction for excluding a component linearly correlated with an element constituting a feature amount of the user's psychological information from the prediction result.

4. A prediction program for causing a computer to execute: a step of receiving an input of attribute information indicating a user's attributes and psychological information indicating the user's psychological state; a step of predicting the user's psychological index using a prediction model learned to output a prediction result of the user's psychological index with the user's attribute information and psychological information as inputs; and a step of performing and outputting a correction for excluding a component linearly correlated with an element constituting a feature amount of the user's psychological information from the prediction result.

Images