JP2020160551A - Analysis support device for personnel item, analysis support method, program, and recording medium - Google Patents

Abstract

To provide a system capable of easily performing adaptive selection and allowing adaptive personnel item analysis when selecting a learning model for the personnel item analysis.SOLUTION: An analysis support device for personnel items of the present invention includes: an input unit for inputting user information; a learning model generation unit that performs learning twice or more using a part of learning data when past personnel data is input as the learning data for an analysis target item related to personnel, and generates two or more learning models to analyze the analysis target item; a selection information generation unit that performs predictive verification for each learning model using the rest of the learning data, and generates an error between true information and prediction information, a reproduction rate, an adaptation rate, and a correct answer rate as selection information for selecting the learning model; an output unit that outputs the selection information for each learning model; and an analysis unit that analyzes the analysis target item using personnel data of an analysis target person using the selected learning model when a selection result of a user and the personnel data of the analysis target person are input.SELECTED DRAWING: Figure 1

Description

The present invention relates to an analysis support device for human resources, an analysis support method, a program, and a recording medium.

In a company, the person in charge of the human resources department usually assigns employees to appropriate departments, proposes training to obtain the desired effect, extracts risks such as retirement, etc. based on a large amount of personnel data. ing. However, such judgments on personnel relations cannot be made objectively, such as strongly reflecting the intentions of specific employees such as superiors, and the larger the company, the greater the amount of information to be referred to. Will be required.

On the other hand, in recent years, in various fields, it has been attempted to make a machine learn about the processing performed by humans, construct a learning model, and process using the learning model. However, since the learning model is usually constructed by an expert and delivered to the user, the user must try using the delivered learning model to determine whether or not the learning data is suitable. I can't judge. Further, regarding the generation of the learning model, a versatile system is on the market, but even if the user generates the learning model using the system, the generated learning model is the purpose as a result. It is difficult to judge whether or not it matches.

Therefore, the present invention provides a system that enables the user to easily make a suitable selection when selecting a learning model in the analysis of the human resources, thereby enabling the analysis of the suitable human resources. With the goal.

In order to achieve the above object, the analysis support device for human matters of the present invention
Including input unit, learning model generation unit, selection information generation unit, output unit, and analysis unit,
The input unit is
Enter the user's information and
The learning model generation unit
When the past personnel data is input as learning data for the analysis target items related to personnel by the input unit,
Two or more learnings are performed using a part of the training data, and two or more learning models to analyze the analysis target item are generated.
The selection information generation unit
Using the rest of the training data, prediction verification is performed for each of the training models, and the error between the true information and the prediction information, the recall rate, the precision rate, and the correct answer rate are determined as selection information for selecting the training model. Generated as
The output unit
For each of the learning models, the selection information is output.
The analysis unit
When the selection result of the user and the personnel data of the analysis target person are input by the input unit,
It is characterized in that the analysis target item is analyzed by using the personnel data of the analysis target person by the selected learning model.

The analysis support method for human resources of the present invention includes an input process, a learning model generation process, a selection information generation process, an output process, and an analysis process.
The input process is
Enter the user's information and
The learning model generation step is
When past personnel data is input as learning data for the analysis target items related to personnel by the input process,
Two or more learnings are performed using a part of the training data, and two or more learning models to analyze the analysis target item are generated.
The selection information generation step is
Using the rest of the training data, prediction verification is performed for each of the training models, and the error between the true information and the prediction information, the recall rate, the precision rate, and the correct answer rate are determined as selection information for selecting the training model. Generated as
The output process is
For each of the learning models, the selection information is output.
The analysis step is
When the selection result of the user and the personnel data of the analysis target person are input by the input process,
It is characterized in that the analysis target item is analyzed by using the personnel data of the analysis target person by the selected learning model.

The program of the present invention is characterized in that a computer executes the analysis support method for the human resources of the present invention.

The recording medium of the present invention is a computer-readable recording medium on which the program of the present invention is recorded.

According to the present invention, for example, a plurality of learning models are generated based on the learning data, and selection information for selecting the learning model is generated for each learning model. Therefore, the user can use the selection information. Based on the above, it is possible to easily select a learning model according to the purpose. Therefore, by using the selected learning model, it is possible to further obtain the analysis result according to the purpose.

FIG. 1 is a block diagram showing an example of the analysis support device of the first embodiment. FIG. 2 is a block diagram showing an example of the hardware configuration of the analysis support device of the first embodiment. FIG. 3 is an example of a flowchart showing the process of the analysis support method of the first embodiment. FIG. 4 is a graph showing an example of guide information in selecting a learning model in the second embodiment. FIG. 5 is a graph showing guide information for setting the threshold value of the selected learning model in the third embodiment.

In the present invention, the user is a group that analyzes personnel items, a person in charge of personnel, or the like. The unit of the "group" is not particularly limited, and may be, for example, an organizational unit such as a company, a union, a school, or an administration, or a department unit such as a human resources department.

According to the analysis support device and the analysis support method for the personnel items of the present invention, it is possible to automatically analyze the personnel items of the person to be analyzed. The personnel items are not particularly limited, and are, for example, proper placement, training and its effects, or withdrawal possibility. “Appropriate placement” means, for example, what kind of department, project, job title, post, etc. is appropriate to be placed. “Training and its effects” means, for example, what kind of training should be provided and what kind of effect can be obtained. The “possibility of withdrawal” means, for example, the possibility of withdrawal from the group to which the employee belongs, and specifically, includes withdrawal from a department, job title, project, etc. in addition to withdrawal (retirement).

An embodiment of the present invention will be described. The present invention is not limited to the following embodiments. In each of the following figures, the same parts are designated by the same reference numerals. Further, the explanations of the respective embodiments can be referred to each other's explanations unless otherwise specified. Further, the configurations of the respective embodiments can be combined unless otherwise specified.

[Embodiment 1]
FIG. 1 is a block diagram showing a configuration of an example of the analysis support device 10 for the human resources of the present embodiment. As shown in FIG. 1A, the analysis support device 10 has an input unit 12, a learning model generation unit 13, a selection information generation unit 14, an output unit 15, and an analysis unit 16. The analysis support device 10 may further include, for example, a storage unit 12. The analysis support device 10 is also referred to as, for example, an analysis support system. The analysis support device 10 may be, for example, one device including the above-mentioned parts, or a device in which the above-mentioned parts can be connected via a communication network. The communication network is, for example, the same as the examples described later. In the analysis support device 10, for example, the processing of each part may be performed on the cloud.

As shown in FIG. 1B, for example, the analysis support device 10 can be connected to the external device 30 via the communication network 20. Examples of the external device 30 include a server, a PC (personal computer), and a terminal such as a tablet, a smartphone, and a mobile phone. In the present embodiment, the user uses the external device 30, and the external device 30 and the analysis support device 10 are connected to each other via the communication network 20. However, the present embodiment is not limited to this. The analysis support device 10 may be, for example, a device (for example, a PC or a server) directly used by the user.

The communication line network 20 is not particularly limited, and a known communication line network can be used, and may be wired or wireless. Examples of the communication line network include an Internet line, a telephone line, a LAN (Local Area Network), WiFi (Wireless Fidelity), and the like.

Next, FIG. 2 illustrates a block diagram of the hardware configuration of the analysis support device 10. The analysis support device 10 includes, for example, a CPU (central processing unit) 101, a memory 102, a bus 103, an input device 104, a display 105, a communication device 106, a storage device 107, and the like. Each part of the analysis support device 10 is connected to each other by an interface (I / F) via a bus 103.

The CPU 101 is a processor that controls the entire analysis support device 10, and is not limited to the CPU, and may be another processor. In the analysis support device 10, for example, the program of the present invention and other programs are executed by the CPU 101, and various information is read and written.

The analysis support device 10 can be connected to the communication network by, for example, the communication device 106 connected to the bus 103, and can also be connected to an external device via the communication network. The external device is not particularly limited, and is, for example, a database, a PC (personal computer), a terminal, or the like, and the terminal is, for example, a tablet, a smartphone, a mobile phone, or the like. The connection method between the analysis support device 10 and the external device is not particularly limited, and may be, for example, a wired connection or a wireless connection. The wired connection may be, for example, a cord connection or a cable connection for using the communication network. The wireless connection may be, for example, a connection using a communication network or a connection using wireless communication. The communication line network is not particularly limited, and for example, a known communication line network can be used, which is the same as described above.

The memory 102 includes, for example, a main memory, and the main memory is also referred to as a main storage device. When the CPU 101 performs processing, for example, the memory 102 reads various operation programs 108 such as the program of the present invention stored in the auxiliary storage device described later, and the CPU 101 receives data from the memory 102. , Program 108 is executed. The main memory is, for example, a RAM (random access memory). The memory 102 further includes, for example, a ROM (read-only memory).

The storage device 107 is also referred to as a so-called auxiliary storage device with respect to the main memory (main storage device), for example. The storage device 107 includes, for example, a storage medium and a drive for reading and writing to the storage medium. The storage medium is not particularly limited, and may be an internal type or an external type, for example, HD (hard disk), FD (floppy (registered trademark) disk), CD-ROM, CD-R, CD-RW, MO, Examples thereof include a DVD, a flash memory, a memory card, and the like, and the drive is not particularly limited. The storage device 107 may be, for example, a hard disk drive (HDD) in which a storage medium and a drive are integrated. For example, as described above, the program 108 is stored in the storage device 107, and as described above, when the CPU 101 is executed, the memory 102 reads the operation program 108 from the storage device 107. Further, the storage device 107 includes, for example, a storage unit 11.

The analysis support device 10 may further include, for example, an input device 104 and a display 105. The input device 104 is, for example, a scanner, a touch panel, a keyboard, or the like. Examples of the display 105 include an LED display and a liquid crystal display.

The input unit 12 inputs user information. The storage unit 11 stores, for example, the information input by the input unit 12. The input of information to the analysis support device 10 is not particularly limited. For example, when the user directly uses the analysis support device 10, the information may be directly input using the input device 104 shown in FIG. When using the external device 30 as shown in FIG. 1 (B), information may be input to the external device 30 and transmitted from the external device 30 to the analysis support device 10 via the communication network 20. ..

The user information is, for example, the past personnel data accumulated by the user in the generation of the learning model, and the personnel data of the analysis target person in the analysis of the analysis target person.

When the past personnel data is input as learning data for the analysis target item related to personnel by the input unit 12, the learning model generation unit 13 performs learning two or more times using a part of the learning data, and the above-mentioned Generate two or more learning models to analyze the items to be analyzed. As the learning model generation unit 13, for example, an existing learning model generation system can be used.

The selection information generation unit 14 performs predictive verification for each of the learning models using the rest of the learning data, and determines the error, recall rate, precision rate, and correct answer rate between the true information and the predicted information as the learning model. Is generated as selection information for selecting.

The output unit 15 outputs the selection information for each of the learning models. The output of the selection information of the analysis support device 10 is not particularly limited. For example, when the user directly uses the analysis support device 10, the selection information may be output and displayed on the display 105 shown in FIG. Alternatively, when the user uses the external device 30 as shown in FIG. 1 (B), the analysis support device 10 outputs the selection information to the external device 30 via the communication network 20 and outputs the selection information to the external device 30. The selection information may be displayed on the display of.

The user can select one of the learning models from the plurality of learning models based on the output selection information, and input the selection result to the analysis support device 10 by the input unit 12 as described above. .. Further, when the user selects the learning model, the user can further input the personnel data of the analysis target person into the analysis support device 10 by the input unit 12 as described above. The personnel data of the person to be analyzed may be input in advance, stored in the storage unit 11, and recalled from the storage unit 11 when performing analysis by the selected learning model.

When the selection result of the user and the personnel data of the analysis target person are input by the input unit 12, the analysis unit 16 uses the personnel data of the analysis target person according to the selected learning model. Analyze the items to be analyzed. This analysis result can be output to the user by, for example, the output unit 15 in the same manner as described above.

Next, the analysis support method of the present embodiment will be described. The analysis support method of the present embodiment includes an input step, a learning model generation step, a selection information generation step, a selection information display step, and an analysis step. The analysis support method of the present embodiment can be carried out using, for example, the analysis support device 10 shown in FIGS. 1 and 2. The analysis support method of the present embodiment is not limited to the use of the analysis support device 10.

The input step is a step of inputting user information, and can be executed by, for example, the input unit 12 of the analysis support device 10.

In the learning model generation step, when past personnel data is input as learning data for analysis target items related to personnel by the input process, a part of the learning data is used to perform learning two or more times. This is a process of generating two or more learning models for analyzing the items to be analyzed. The step can be executed, for example, by the learning model generation unit 13 of the analysis support device 10.

In the selection information generation step, prediction verification is performed for each of the training models using the rest of the training data, and the error, recall rate, precision rate, and correct answer rate between the true information and the prediction information are determined by the learning model. It is a process of generating as selection information for selecting. The step can be executed, for example, by the selection information generation unit 14 of the analysis support device 10.

The output step is a step of outputting the selection information (selection information output step) for each of the learning models, and can be executed by, for example, the output unit 15 of the analysis support device 10.

In the analysis step, when the selection result of the user and the personnel data of the analysis target person are input by the input step, the personnel data of the analysis target person is used by the selected learning model. This is the process of analyzing the items to be analyzed. The step can be executed, for example, by the analysis unit 16 of the analysis support device 10. The analysis support method of the present embodiment further includes, for example, an output step (analysis result output step), and the analysis result can be output to the user.

Next, the analysis support method using the analysis support device 10 of the present embodiment will be described with reference to the flowchart of FIG. 3, taking as an example the analysis of the proper arrangement of the analysis target persons.

First, the input unit 12 inputs past personnel data regarding proper placement (S101). The personnel data is, for example, information on the members constituting the group, and job-related information on the individual members. The items of the job-related information include, for example, job history (for example, transfer history, post history), skill history, performance evaluation, etc., and also include, for example, possession qualification, award history, career plan, etc. It may be. Moreover, if the items are related to each other, the relationship may be further linked. Since the personnel data is processed by the analysis support device 10, for example, the data is quantified in advance, and the quantification method is not particularly limited, and a general method in learning model generation can be adopted.

Next, the learning model generation unit 13 performs learning a plurality of times using a part of the input personnel data to generate a plurality of learning models (S102). Specifically, one learning model is generated for each learning. The personnel data is randomly divided into two groups, one of which is used to generate a learning model, and the rest is used for verification described later. The ratio of the personnel data used to generate the learning model is not particularly limited, and the lower limit is, for example, 70% or more, the upper limit is, for example, 80% or less, and as a specific example, 80%. Use the personnel data of. Here, "wari" means the ratio of the number of people in the personnel data to be used to the total number of people included in the personnel data, and may be expressed in%.

Then, the selection information generation unit 14 performs predictive verification for each of the learning models using the rest of the learning data (S103), and from the true information of the personnel data and the predictive information of the predictive verification, both The error, recall rate, accuracy rate, and correct answer rate are generated as selection information for selecting a learning model (S104). In the prediction verification, for example, whether or not the true information and the prediction information match can be set with an arbitrary threshold value (for example, 0.6), and may be changed as appropriate.

For example, the smaller the error between the true information and the predicted information, the better the learning model, and the larger the error, the worse the learning model. The method for calculating the error is not particularly limited, and the methods for calculating the root mean square error (RMSE) and the average absolute error (MAE) shown in the following equation can be used. In the following formula, n is the number of personnel data, y ₁ ... y _n is true information (true value), and f ₁ ... f _n is predicted information (predicted value).

The recall rate, the precision rate, and the correct answer rate can be calculated using, for example, the result of prediction verification. According to the prediction verification, for example, the number of data of TP, TN, FP, and FN shown below can be obtained. Table 1 below shows an example in which the classification results of the number of data of TP, TN, FP, and FN are summarized as a confusion matrix.
"TP" The number of data whose true information is positive (with appropriateness) and the prediction result is positive (with appropriateness) "TN" The number of data whose true information is negative (not appropriate) and whose prediction result is negative (not appropriate) "FP" Number of data with true information negative (not appropriate) and prediction result positive (with appropriate) "FN" Number of data with true information positive (with appropriate) and prediction result negative (not appropriate)

The recall (Recall) is the ratio of the data predicted to be “appropriate” among the personnel data whose true information is “appropriate”, and can be calculated by, for example, the following formula. In the case of Table 1, 1985 / (1985 + 167) = 0.922 (92.2%).
Recall = TP / (TP + FN)

The precision is the ratio of personnel data whose true information is "appropriate" among the data predicted to be "appropriate", and can be calculated by, for example, the following formula. In the case of Table 1, 1985 / (1985 + 128) = 0.939 (93.9%).
Precision = TP / (TP + FP)

The accuracy rate is data in which the true information is "appropriate" and the prediction information is "appropriate", and the true information is "appropriate" and the prediction information is "appropriate". The total with the data is the ratio of the total data, and can be calculated by, for example, the following formula. In the case of Table 1, (1985 + 11720) / (1985 + 128 + 11720 + 167) = 0.979 (97.9%).
Accuracy = (TP + TN) / (TP + FP + TN + FN)

Then, the output unit 15 outputs the selection information for each learning model (S105). The selection information is output in association with, for example, the identification information (for example, a number) of the learning model, and is displayed on the display of the analysis support device 10 or the external device 30. An example of the selection information displayed on the display is shown in Table 2 below.

Based on the output selection information of the learning model, the user can arbitrarily select a desired learning model from the plurality of the learning models. As the selection information, for example, as shown in Table 2, the RMSE and MAE indicating the error, the recall rate, the precision rate, and the correct answer rate are displayed. Therefore, if the numerical values of these items are taken into consideration, the user Can select the desired learning model. As described above, the RMSE and MAE can be judged to be learning models with less error as the numerical values are smaller, and the recall rate, precision rate and correct answer rate are judged to be more reliable as the numerical values are larger. can do. As described above, according to the present embodiment, since the information for selecting the learning model is output, the user himself / herself can easily select the learning model based on the information.

Then, the input unit 12 inputs the selection result of the user, that is, which learning model to use (S106), and further, when the personnel data of the analysis target person is input (S107), the analysis unit 16 (S108), the selected learning model is used to analyze the analysis target item using the personnel data of the analysis target person. Then, the obtained analysis result is output to the user by the output unit 15 (S109), and ends (END). The selection result of the user can be input, for example, by selecting (converting to a black circle) the corresponding part of the selection field on the display as shown in Table 2.

[Embodiment 2]
In the selection information output step (S105) of the first embodiment, for example, the following information may be output as supplementary information for the user to select one learning model from a plurality of learning models. ..

FIG. 4A shows a graph of a curve showing the relationship between the precision rate and the recall rate. According to this curve, for example, as the area under the curve (AUC: Area Under the Curve) is larger, there is a point (position at the upper right of the graph) where both the precision rate and the recall rate are higher, and the AUC. The larger the value, the better the learning model. Therefore, for example, in the step (S105), the user can select a learning model more by outputting guide information that the larger the AUC is, the better the learning model is, in accordance with this graph. Becomes easier.

FIG. 4B shows a graph of a curve showing the relationship between the true positive rate and the false negative rate. The true positive rate is the ratio at which the true information of the objective variable (appropriateness) can be predicted to be "appropriate" in the data of "appropriate", and the false negative rate is the objective variable (appropriateness). The true information of) indicates the percentage of the data that is "not appropriate" that is predicted to be "appropriate". FIG. 4B shows the relationship between the true positive rate and the false negative rate when the threshold value of the objective variable (appropriateness) is changed. According to this curve, for example, the larger the AUC, the higher the true positive rate and the false negative rate (at the upper left position of the graph), and the larger the AUC, the relatively. It's a good learning model. Therefore, for example, in the step (S105), the user can select a learning model more by outputting guide information that the larger the AUC is, the better the learning model is, in accordance with this graph. Becomes easier.

FIG. 4C shows a graph showing the relationship between the true (actual) objective variable (appropriateness) or the prediction objective variable (appropriateness) and the actual objective variable (appropriateness). In this graph, the true objective variable (appropriateness) is always a straight line, and the deviation from the prediction objective variable (appropriateness) represents the deviation between the true information and the prediction information. Therefore, according to this graph, as a result of the prediction verification, it is visually determined how much the prediction objective variable (appropriateness) deviates from the straight line of the true objective variable (appropriateness). can do. Therefore, for example, in the step (S105), the smaller the dissociation of the prediction objective variable (appropriateness) with respect to the straight line of the true objective variable (appropriateness), the better. By outputting the guide information that the learning model is used, the user can more easily select the learning model.

[Embodiment 3]
After the output step (S109) of the analysis result of the analysis target person using the learning model of the first embodiment, the output unit 15 further outputs, for example, guide information regarding the setting of the threshold value for the analysis of the analysis target person. You may.

FIG. 5 shows a graph showing the relationship between the precision and the recall rate and the threshold value. According to this curve, for example, when the threshold value of the objective variable (appropriateness), that is, the branch point of the objective variable (appropriateness) that separates "appropriate" and "not appropriate" is set to "0.6 or less". The recall rate is close to 100%, but when the threshold value is "exceeding 0.6", the recall rate decreases, and when the threshold value is about 0.65, it decreases to about 50%. On the other hand, regarding the suitability, when the threshold value is "exceeding 0.6", the value is close to 100%, but when the threshold value is "0.6 or less", the suitability is lowered. At a threshold of about 0.56, it drops to about 64%. Then, for example, when analyzing the appropriate placement, the priority of the recall rate has the meaning of giving priority not to overlook the person who has the appropriateness, and the priority of the precision rate mistakenly "appropriately" the person who does not have the appropriateness. It has the meaning of giving priority to not saying "yes". Therefore, for example, after the step (S109), by outputting the meaning of the priority of the recall rate and the priority of the precision rate as guide information in accordance with this graph, the user can set the threshold value according to the purpose. Can be set.

[Embodiment 4]
The program of the present embodiment is a program capable of executing the analysis support method of the present invention on a computer. Alternatively, the program of this embodiment may be recorded on, for example, a computer-readable recording medium. The recording medium is not particularly limited, and examples thereof include the storage medium described above.

Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made within the scope of the present invention in terms of the structure and details of the present invention.

According to the present invention, for example, a plurality of learning models are generated based on the learning data, and selection information for selecting the learning model is generated for each learning model. Therefore, the user can use the selection information. Based on the above, it is possible to easily select a learning model according to the purpose. Therefore, by using the selected learning model, it is possible to further obtain the analysis result according to the purpose.

10 Analysis support device 11 Storage unit 12 Input unit 13 Learning model generation unit 14 Selection information generation unit 15 Output unit 16 Analysis unit 20 Communication network 30 External device

Claims (8)

Including input unit, learning model generation unit, selection information generation unit, output unit, and analysis unit,
The input unit is
Enter the user's information and
The learning model generation unit
When the past personnel data is input as learning data for the analysis target items related to personnel by the input unit,
Two or more learnings are performed using a part of the training data, and two or more learning models to analyze the analysis target item are generated.
The selection information generation unit
Using the rest of the training data, prediction verification is performed for each of the training models, and the error between the true information and the prediction information, the recall rate, the precision rate, and the correct answer rate are determined as selection information for selecting the training model. Generated as
The output unit
For each of the learning models, the selection information is output.
The analysis unit
When the selection result of the user and the personnel data of the analysis target person are input by the input unit,
An analysis support device for a person's item, which analyzes the analysis target item using the personnel data of the analysis target person by the selected learning model. The analysis support device according to claim 1, wherein the analysis target item is proper placement, training and its effect, or withdrawal possibility. In addition, including the output section
The analysis support device according to claim 1 or 2, wherein the output unit outputs the analysis result of the analysis unit. Includes input process, learning model generation process, selection information generation process, output process, and analysis process.
The input process is
Enter the user's information and
The learning model generation step is
When past personnel data is input as learning data for the analysis target items related to personnel by the input process,
Two or more learnings are performed using a part of the training data, and two or more learning models to analyze the analysis target item are generated.
The selection information generation step is
Using the rest of the training data, prediction verification is performed for each of the training models, and the error between the true information and the prediction information, the recall rate, the precision rate, and the correct answer rate are determined as selection information for selecting the training model. Generated as
The output process is
For each of the learning models, the selection information is output.
The analysis step is
When the selection result of the user and the personnel data of the analysis target person are input by the input process,
A method for supporting analysis of personnel items, which analyzes the items to be analyzed by using the personnel data of the person to be analyzed by the selected learning model. The analysis support method according to claim 4, wherein the analysis target item is proper placement, training and its effect, or withdrawal possibility. In addition, it includes an output process
The analysis support method according to claim 4 or 5, wherein the output step outputs an analysis result of the analysis step. A program characterized by causing a computer to execute the analysis support method for a human resource item according to any one of claims 4 to 6. A computer-readable recording medium on which the program according to claim 7 is recorded.