JP6762584B2 - Learning model construction device, post-employment evaluation prediction device, learning model construction method and post-employment evaluation prediction method - Google Patents

Description

The present invention is a learning model construction device for predicting post-employment evaluation of a prospective employee in a company, a post-employment evaluation prediction device using a learning model constructed by the learning model construction device, and a post-employment evaluation prediction device for a prospective employee in a company. The present invention relates to a learning model construction method for predicting evaluation and a post-employment evaluation prediction method using a learning model constructed by the learning model construction device.

In order to select a person who is suitable for the company from a wide range of human resources during recruitment activities such as new graduate recruitment and mid-career recruitment, the company not only conducts an interview test but also provides the duties provided by the company. There are cases where aptitude tests and job aptitude tests developed by the company are conducted. While having all applicants take such job aptitude tests uniformly can be the basis of objective evaluation, there are also aspects that do not match the corporate culture and required abilities because it is uniform. It has been pointed out. In order to deal with this point, there is a technology that attempts to evaluate the results of the job aptitude test for applicants who want to join the company based on criteria that are linked to the internal evaluation as well as the job aptitude test for employees who actually work at the company (for example). (See Patent Document 1)

Japanese Patent No. 5171437

However, the technology disclosed in Patent Document 1 refers to the results of aptitude tests that employees who have already received high in-house evaluations in a company have taken in the past, and applicants for joining the company based on the results of job aptitude tests of highly evaluated employees. (Refer to paragraph 0007 of Patent Document 1), on the other hand, using the average value of highly evaluated employees has various characteristics for companies. There is a problem that it does not consider that there are highly evaluated human resources, and it is not possible to guess the reason why the employee who joined the company leaves the job. Although it is a rough guess, in light of the recent trend, it is said that most new graduates have undergone some kind of job aptitude test and the hiring side also attaches great importance to this, but according to the announcement by the Ministry of Health, Labor and Welfare, 3 of the new graduates There is a fact that over-the-counter employees have left their jobs within three years after getting a job. Looking at this situation from the corporate side, the problem is that human resources who have been trained at great expense after hiring will eventually leave their jobs.

Therefore, an object of the present invention is to enable the hiring activity to be carried out while predicting the post-employment evaluation of the human resources to be hired from various aspects, and to eliminate the mismatch that leads to the turnover as much as possible.

In order to achieve the above object, the present invention includes the following configurations. The definitions of terms used to describe the invention described in any of the claims shall be applied to other inventions to the extent possible due to their nature, regardless of the category or order of description of the invention.

(Invention according to claim 1)
The learning model construction device according to the invention according to claim 1 (hereinafter referred to as "construction device according to claim 1") is a current or past employee (hereinafter referred to as "adopting company") of a company scheduled to be adopted (hereinafter referred to as "adopting company"). (Hereinafter referred to as "existing employees") Examination data acquisition means for acquiring the results of taking the prescribed job aptitude test as test data, and evaluation data acquisition for acquiring evaluation data showing the results of internal evaluations of each existing employee. A means and a label acquisition means for acquiring the evaluation data acquired by the evaluation data acquisition means divided into three levels of H (high), M (intermediate), and L (low) as each evaluation label. , Hope to join the hiring company by conducting supervised learning using each set of test data acquired by the test data acquisition means and each set of labels acquired by the corresponding label acquisition means as teacher data. It is equipped with a learning means for constructing a learning model for predicting post-employment evaluation of applicants who have taken the appropriate test, and the evaluation data is data that negatively evaluates the turnover of the existing employee. It is a feature.

According to the construction device of claim 1, a learning model based on current or past employee test data and evaluation data can be constructed. Since it is premised on building not only current employees but also past employees, it is possible to build a learning model that can predict not only the positive activity of applicants for joining the company but also the negative activity including turnover. .. The reason for including turnover is that turnover can be interpreted as having a negative job performance.

(Invention according to claim 2)
The learning model construction device according to the invention according to claim 2 (hereinafter, referred to as “construction device according to claim 2”) is an individual including the history and work history of each of the existing employees, as a preferred embodiment of the construction device according to claim 1. Further provided with personal history data acquisition means to be acquired as career data, the learning means supervised learning using the group obtained by adding the personal history data of each of the existing employees acquired by the personal history data acquisition means as teacher data. It is characterized in that it is configured to do so.

According to the construction device of claim 2, it is possible to provide a learning model that can predict the degree of prediction of the evaluation of a prospective employee in more detail. By using supervised learning, a learning model is constructed by finding a suitable combination of test data for predicting post-employment evaluation from a large number of test data combinations without the need for prior human experience or intuition. be able to. In other words, by taking into account the history of current or past employees until they are hired by the company and the career history of the work they experienced after joining the company, for example, the evaluation in the sales position and the evaluation in the general affairs position can be evaluated respectively. It can make it easier to build a predictable learning model.

(Invention according to claim 3)
The learning model construction device according to the invention according to claim 3 (hereinafter, referred to as “construction device according to claim 3”) is a preferred embodiment of the construction device according to claim 1 or 2, wherein the learning means teaches the teacher data. It is configured to construct a plurality of different learning models by performing a plurality of different supervised learning as data, and further, a test acquired by the test data acquisition means for each of the plurality of learning models constructed by the learning means. It is characterized in that a matching degree determining means for determining the matching degree between each of the test labels output by inputting the data and each of the evaluation labels acquired by the label acquiring means is provided.

According to the construction device of claim 3, different learning models are obtained (except for coincidence) by performing each of the different supervised learnings. Here, each test label and each evaluation label can be compared, and the degree of conformity can be determined. Algorithms used in supervised learning, such as "decision trees" and "random forests," have been found to vary in accuracy depending on the type and number of data used. In order to compensate for the decrease in accuracy due to this difference, it is meaningful to search for a more accurate learning model by using a method in which the type and number of data used are changed for each algorithm. As a result of the search, it becomes possible to select a learning model related to the evaluation label having the highest degree of matching, that is, the evaluation label that can appropriately predict the evaluation after joining the company according to the actual situation of the company.

(Invention according to claim 4)
The learning model construction device according to the invention according to claim 4 (hereinafter, referred to as “construction device according to claim 4”) is different from the adopted company as a preferred embodiment of the construction device according to any one of claims 1 to 3. Or, the current or past employees of two or more other companies (hereinafter referred to as "existing employees of other companies") have acquired the results of taking the prescribed job aptitude test as other company's test data, and the other company's existing test data acquisition method. The label acquisition means further includes another company's evaluation data acquisition means for acquiring other company's evaluation data showing the result of each employee's in-house evaluation, and the label acquisition means uses the evaluation data (in-house evaluation data) acquired by the evaluation data acquisition means for the other company. It is characterized in that each of the merged data, which is a combination of the evaluation data of other companies acquired by the evaluation data acquisition means, is divided into a plurality of levels and acquired as each evaluation label.

According to the construction device of claim 4, it is possible to construct a learning model of the hiring company while supplementing the in-house evaluation data and the like with the evaluation data of other companies. Since it is premised on building not only current employees but also past employees in the company and other companies, a learning model that can predict not only the positive activity of applicants for joining the company but also the negative activity including turnover. Can be built.

(Invention according to claim 5)
The learning model construction device according to the invention according to claim 5 (hereinafter, referred to as "construction device according to claim 5") is a current or past employee of one or more other companies (hereinafter, "" (Referred to as "existing employees of other companies") Acquire the means of acquiring other company's test data, which acquires the results of taking the prescribed job aptitude test as other company's test data, and the other company's evaluation data, which shows the results of each other company's internal evaluation. Each evaluation label is divided into three levels, H (high), M (intermediate), and L (low) , for each of the other company's evaluation data acquisition means and the other company's evaluation data acquired by the other company's evaluation data acquisition means. Perform supervised learning by using the other company's label acquisition means to be acquired, the other company's test data acquired by the other company's test data acquisition means, and the corresponding sets of labels acquired by the other company's label acquisition means as other company's teacher data. Therefore, it is characterized by providing other companies' learning means for constructing a learning model for post-employment evaluation prediction of a person who wishes to join the hiring company and has taken the aptitude test.

According to the construction device of claim 5, a learning model of the hiring company can be constructed by utilizing the evaluation data of other companies regarding the employees of other companies. Since it is premised on building not only current employees but also past employees at other companies, we will build a learning model that can predict not only the positive activity of applicants for joining the company but also the negative activity including turnover. be able to.

(Invention according to claim 6)
The post-employment evaluation prediction device according to the invention of claim 6 (hereinafter, referred to as “prediction device of claim 6”) uses a learning model constructed by any of the construction devices of claims 1, 2, 4, and 5. A preferred embodiment of the post-employment evaluation prediction device is an applicant test data acquisition means for acquiring the result of the applicant for the job aptitude test as the applicant examination data, and the applicant test data acquisition means. It is characterized by comprising an evaluation prediction means for predicting the post-employment evaluation of the applicant for joining the company based on each of the acquired applicant test data of the applicant for joining the company and the learning model.

According to the prediction device of claim 6, the degree of activity of a person who wishes to join the company after joining the company can be predicted with high accuracy including the possibility of leaving the job. In particular, when the learning model constructed by the device of claim 2 is used, personal history data including the history and work history of each existing employee is also taken into consideration. Therefore, for example, the type of business, business type, and size of the company. Furthermore, since differences in departments within the company and the environment of bosses and colleagues are also reflected, it is possible to provide a learning model that is more suitable for the actual situation of the company.

(Invention according to claim 7)
The post-employment evaluation prediction device according to the invention according to claim 7 (hereinafter, referred to as “prediction device according to claim 7”) is a post-employment evaluation using a plurality of learning models constructed by the learning model construction device according to claim 3. A preferred embodiment of the prediction device, which is a means of acquiring the applicant test data for acquiring the result of the applicant for joining the company taking the predetermined job suitability test as the applicant test data, and a matching degree determined by the matching degree determining means. Based on the selection means for selecting the learning model with the highest degree of matching, the applicant test data of the applicant for joining the company acquired by the applicant examination data acquisition means, and the learning model, the applicant for joining the company It is characterized by being provided with an evaluation prediction means for predicting evaluation after joining the company.

According to the invention of claim 7, it is possible to predict the post-employment evaluation of a person who wishes to join the company with high accuracy by using a learning model that conforms to the actual situation of the company among different learning models. Needless to say, the possibility of leaving a job can be predicted with high accuracy in light of the learning model.

(Invention according to claim 8)
The learning model construction method according to the invention according to claim 8 (hereinafter, referred to as "construction method of claim 8") is a post-employment evaluation prediction method executed by a computer, and is a company planning to adopt it (hereinafter,). , The test data acquisition step to acquire the result of each current or past employee (hereinafter referred to as "existing employee") of the "employed company" as the test data, and the in-house of each existing employee The evaluation data acquisition step for acquiring the evaluation data showing the evaluation result and the evaluation data acquired in the evaluation data acquisition step are divided into three levels of H (high), M (intermediate), and L (low). , The label acquisition step acquired as each evaluation label, each test data acquired in the test data acquisition step, and each set of labels acquired in the corresponding label acquisition step are used as teacher data for supervised learning. By doing so, the evaluation data is provided with a learning step for constructing a learning model for post-employment evaluation prediction of a person who wishes to join the hiring company and has taken the aptitude test. It is characterized by the fact that the data is a negative evaluation of employee turnover.

According to the construction method of claim 8, a learning model based on current or past employee test data and evaluation data can be constructed. By using supervised learning, a learning model is constructed by finding a suitable combination of test data for predicting post-employment evaluation from a large number of test data combinations without the need for prior human experience or intuition. be able to. In addition, since it is premised on building not only current employees but also past employees, we will build a learning model that can predict not only the positive activity of applicants for joining the company but also the negative activity including turnover. be able to. The reason for including turnover is that turnover can be interpreted as having a negative job performance.

(Invention according to claim 9)
The learning model construction method according to the invention according to claim 9 (hereinafter, referred to as "construction method of claim 7") is a preferred embodiment of the construction method of claim 8, and is an individual including the history and work history of each of the existing employees. A step for acquiring personal history data to be acquired as career data is further provided, and in the learning step, supervised learning is performed using the group obtained by adding the personal history data of each of the existing employees acquired in the personal history data acquisition step as teacher data. It is characterized by doing.

According to the construction method of claim 9, it is possible to provide a learning model that can predict the degree of prediction of a prospective employee in more detail. In other words, by taking into account the history of current or past employees until they were hired by the company and the career history of the duties they experienced after joining the company, for example, the degree of activity in the sales position and the degree of activity in the general affairs position can be determined. You can build a predictable learning model.

(Invention according to claim 10)
The learning model construction method according to the invention according to claim 10 (hereinafter, referred to as “prediction method of claim 10”) is a preferred embodiment of the construction method according to claim 8 or 9, in which existing employees of other companies have predetermined duties. The other company's test data acquisition step to acquire the result of taking the aptitude test as the other company's test data and the other company's evaluation data acquisition step to acquire the other company's evaluation data showing the result of the in-house evaluation of each existing employee of the other company are further provided. In the label acquisition step, the merged data obtained by combining the evaluation data acquired by the evaluation data acquisition step with the evaluation data of other companies acquired in the evaluation data acquisition step of the other company is divided into a plurality of levels as each evaluation label. It is characterized by acquiring.

According to the prediction method of claim 10, it is possible to construct a learning model of the hiring company while supplementing the in-house evaluation data and the like with the evaluation data of other companies. Since it is premised on building not only current employees but also past employees in the company and other companies, a learning model that can predict not only the positive activity of applicants for joining the company but also the negative activity including turnover. Can be built.

(Invention according to claim 11)
The learning model construction method according to the invention according to claim 11 (hereinafter, referred to as "construction method of claim 11") is a post-employment evaluation prediction method executed by a computer.
The results of each current or past employee of one or more other companies (hereinafter referred to as "existing employees of other companies"), which is different from the company scheduled to be hired, have taken the prescribed job aptitude test as other company test data. The other company's test data acquisition step to be acquired, the other company's evaluation data acquisition step to acquire the other company's evaluation data showing the results of each other company's internal evaluation, and the other company's evaluation data acquired in the other company's evaluation data acquisition step The other company's label acquisition step, which acquires H (high), M (intermediate), and L (low) as their respective evaluation labels, and the other company's test data acquired in the other company's test data acquisition step, and these After joining the company who wished to join the hiring company and took the aptitude test by performing supervised learning using each set of labels acquired in the other company's label acquisition step corresponding to It is characterized by having a learning step of another company that builds a learning model for evaluation prediction, and the evaluation data is data that negatively evaluates the turnover of the existing employee.

According to the construction method of claim 11, it is possible to build a learning model of the hiring company by utilizing the evaluation data of other companies regarding the employees of other companies. Since it is premised on building not only current employees but also past employees at other companies, we will build a learning model that can predict not only the positive activity of applicants for joining the company but also the negative activity including turnover. be able to.

(Invention according to claim 12)
The learning model construction method according to the invention according to claim 12 (hereinafter, referred to as "construction method according to claim 12") is a preferred embodiment of the construction method according to claim 8 or 9, and in the learning step, the teacher A plurality of different learning models are constructed by performing a plurality of different supervised learning as data, and further, each of the test data acquired by the test data acquisition means is inserted into each of the plurality of learning models constructed in the learning step. It is characterized in that a matching degree determination step for determining the matching degree between each of the test labels output by the above and each of the evaluation labels acquired in the label acquisition step is provided.

According to the construction method of claim 12, different learning models can be obtained (except for coincidence) by performing each of the different supervised learnings. Here, the level matching degree of each learning model can be compared with the level of the test data, and the matching degree can be determined. As a result, it becomes possible to select a learning model with the highest degree of matching, that is, the most appropriate prediction of post-employment evaluation according to the actual situation of the company.

(Invention according to claim 13)
The post-employment evaluation prediction method according to the invention according to claim 13 (hereinafter, referred to as “prediction method according to claim 13”) is a post-employment evaluation using a plurality of learning models constructed by the learning model construction method according to claim 12. This is a prediction method, from the applicant test data acquisition step in which the applicant for joining the company acquires the result of taking the predetermined job aptitude test as the applicant test data, and the matching degree determined in the matching degree determination step. Post-employment evaluation of the applicant based on the selection step to select the learning model with the highest degree of matching, the applicant examination data of the applicant who wants to join the company acquired in the data acquisition step of the applicant, and the learning model. It is characterized by including an evaluation prediction step for predicting.

According to the invention of claim 13, among different learning models, by using a learning model that conforms to the actual situation of the company, it is possible to predict the post-employment evaluation of the applicant for joining the company with high accuracy. Needless to say, the possibility of leaving a job can be predicted with high accuracy in light of the learning model.

According to the present invention, it is possible to carry out recruitment activities while predicting the degree of activity of the human resources to be hired after joining the company, and at the same time, it is possible to eliminate as much as possible a mismatch that leads to turnover.

It is a block diagram of the information processing apparatus which concerns on this invention. It is a block diagram of the information processing apparatus which concerns on this modification. It is a flowchart which shows the processing procedure of the learning model construction and evaluation prediction after joining a company which concerns on this invention. It is a flowchart of this modification corresponding to the flowchart of FIG. It is a chart which shows the proof result which concerns on Example 1 of this invention. It is a chart which shows the proof result which concerns on Example 2 of this invention.

(This Embodiment)
Hereinafter, the best mode for carrying out the present invention (hereinafter, referred to as “the present embodiment”) will be described with reference to the drawings. In the following, the main terms will be defined first, and then the specific forms will be explained.

<Definition>
The "job aptitude test" in this specification refers to a test conducted for the purpose of grasping the ability / appropriateness and personality of a person who wishes to join the company, such as "occupational aptitude test", "occupational aptitude diagnosis", and "appropriate test". Sometimes called. Generally, the result of each inspection item is shown as a deviation value. The content of the job aptitude test is not limited, but publicly known ones provided by various vendors as well as independently developed ones can be used. For job aptitude tests provided by the former contractor, for example, "SPI" (trademark) provided by Recruit Management Solutions Co., Ltd., GAB (trademark), CAB (trademark) provided by SHL-JAPAN Co., Ltd. , Tamatebako (trademark), TG-WEB (trademark) provided by Humanage Co., Ltd., MBTI (trademark) provided by Japan MBTI Association, CUBIC provided by e-Human Resources Co., Ltd., etc. In the present specification, such a job aptitude test provided by a trader is referred to as a “company-provided test”.

The "evaluation" in this specification is basically various evaluations (hereinafter, "evaluation") regardless of the name, such as job evaluation and personnel evaluation performed within the company that is scheduled to be hired (hereinafter referred to as "hiring company"). "In-house evaluation") is included, and generally, the one that is quantified and scored for each item in the company is used. For companies whose in-house evaluation is not quantified or scored, it is recommended to use this after quantifying or scoring. The evaluation range of in-house evaluation is not limited to so-called job performance, job performance, job adaptability, etc., and it is preferable to include, for example, turnover. The reason for including turnover is that turnover can be interpreted as having a negative job performance. In a company, if an existing employee wants to move to another department, or if there are positive or developmental reasons such as gaining experience, for example, the current department does not suit my skin, so I just want to leave the department. It does not prevent you from considering the case where there is a negative / avoidant reason.

In addition to the above-mentioned in-house evaluation or in place of the in-house evaluation, the hiring company shall utilize the same job evaluation, etc. in one or two or more other companies that are different from the hiring company (hereinafter, the evaluation in the other company is referred to as "other company's evaluation"). You can also. In other words, the hiring company uses the combination of its own evaluation and the evaluation of other companies as one evaluation, and uses the evaluation of other companies when there is no undeveloped evaluation itself or when the number of employees is extremely small (diverting). can do. Employers can also process data that users think is necessary for such utilization. The selection of a company of another company is, in principle, left to the practitioner of the present invention, such as the person in charge of the hiring company, without any special conditions. In general, it is preferable to select a company that has characteristics similar to the characteristics of the hiring company (for example, industry, scale, profitability, profit model, human resources to which it belongs, and at least a part of the values of the organization). The reason why it is preferable is that similar characteristics are considered to be more in line with the actual situation of the hiring company.

The “personal career data including history / work history” of existing employees in this specification refers to the history and work history of existing employees, as well as, for example, job type, grade (for example, regular employee, manager, management), and assigned department. (For example, it can be grasped by section unit, department unit, business unit unit), assigned boss (for example, it can be grasped by section manager A, department B, manager 丙 or a combination of these), until the final interview Data (interview value, etc.) and information related to the evaluation score of the hiring interview and the evaluator are applicable. It does not mean that all of these must be included, but some or all of them may be included, nor do they preclude the inclusion of anything other than the above.

The term "supervised learning" as used herein refers to one machine learning in which data given in advance is regarded as teacher's advice and learned by using it as a guide. Preferable examples of supervised learning include decision trees, random forests, neural networks, deep learning, logistic regression, support vector machines, naive bays, boosting, bagging, and unique algorithms built by combining these. Of course, it goes without saying that supervised learning methods other than the above can be freely adopted.

In this specification, "applicants for joining the company" are broadly considered. This includes those who have not yet joined the company who literally want to join the company, as well as those who have not yet been ordered or requested to transfer to another department by existing employees. In other words, the other departments in the latter intra-company transfer fall under the former company. In other words, if a person becomes a member of a new world (company, other department), the purpose is to predict the evaluation of how much he / she can play an active role in that new world.

(Outline configuration of evaluation prediction device)
FIG. 1 is a functional block diagram showing a configuration example of an evaluation prediction device (hereinafter referred to as “prediction device”) for predicting post-employment evaluation of applicants (pre-hiring employees).

As shown in the lower column of FIG. 1, the prediction device 1 is, for example, an information processing device such as a smartphone, a tablet computer, a general-purpose computer, or a workstation, and is not shown because it is incorporated therein. , Central processing unit (CPU), main memory (RAM), read-only memory (ROM), input / output device (I / O), and external storage device such as hard disk device (not shown) as needed. It consists of the equipment that is used. As will be described later, the prediction device 1 also has a function of a learning model construction device (hereinafter, referred to as “construction device”). Further, the prediction device 1 does not have to be composed of one device, but may be composed of a plurality of devices that are communicably connected to each other by a network.

In the above-mentioned ROM or external storage device, a program for making the information processing device function as a prediction device 1 or included in the prediction device 1, a post-employment evaluation prediction method (hereinafter referred to as "prediction method"), or a learning model construction method. A program for causing a computer to execute (hereinafter referred to as "construction method") is stored. When such a program is placed on the main memory and executed by the CPU, the prediction device 1 or the construction device 3 included in the prediction device 1 is realized, and the prediction method and the construction method are executed.

Further, the above program does not necessarily have to be stored in the storage device in the information processing device, and computer-readable program recording such as a magnetic disk, an optical disk, a magneto-optical disk, a CD (Compact Disc), or a DVD (Digital Video Disc). It is provided from a medium, an external device (for example, an ASP (Application Service Provider) server, etc.), cloud computing, or the like, and does not prevent the configuration in which this is placed in the main memory. Hereinafter, a specific description will be given.

The evaluation prediction device 1 shown in FIG. 1 includes a learning model construction means 3 (hereinafter referred to as “construction means 3”), an evaluation prediction means 5 after joining the company (hereinafter referred to as “prediction means 5”), and an external mouse or keyboard. It is basically composed of an input device 7 including the above, and a display device 9 including a liquid crystal display for direct display and a printer for indirect display. The construction means 3 is a device for constructing a learning model that predicts the post-employment evaluation of applicants for joining the company. The prediction means 5 is a device that predicts the post-employment evaluation of a person who wishes to join the company by using the learning model constructed by the construction means 3. The input device 7 is a device for inputting information, and the display device is a device for displaying an expected result or the like. The processing device 1 is configured to communicate with the outside via the network N and exchange information. In this embodiment, the results of the job aptitude test for each existing employee can be imported from the external vendor-provided test database.

(Outline configuration of construction means)
As shown in FIG. 1, the construction means 3 includes a test data acquisition means 31, an evaluation data acquisition means 33, a label acquisition means 35, a learning means 37, and a matching degree determination means 39, and as a preferred embodiment, the personal history data acquisition means is further provided. It has 38. In addition, it has an existing employee examination information database 61, an existing employee evaluation information database 62, a label information database 63, and a learning model database 69, and preferably has an existing employee personal history database 65 and an existing employee interview result database 67. ..

The test data acquisition means 31 includes the existing employee test information database 61, and is a means for each existing employee (that is, an individual existing employee, the same applies hereinafter) to acquire the result of taking a predetermined job aptitude test as test data. is there. In the job aptitude test in the present embodiment, the trader-provided test results are acquired from the trader-provided test result database 101 outside the device and stored in the existing employee test information database 61 in the device so as to be searchable.

The evaluation data acquisition means 33 includes the existing employee evaluation information database 62, and acquires evaluation data showing the results of internal evaluation of each existing employee. The acquired evaluation data is associated with each existing employee and stored in the existing employee evaluation information database 62 in a searchable manner. As defined above, various evaluations are used for this evaluation.

The label acquisition means 35 includes the label information database 63, and is divided into a plurality of levels (for example, H (high), M (intermediate), and L (low)) by scoring the evaluation acquired earlier. For example, the evaluation of the existing employee A obtains "H", the existing employee B obtains "M", and the existing employee C obtains "L" as labels. In light of the rule of thumb, the three-step level is practical, but it may be two-step, four-step, five-step, or even more. The acquired label is associated with each existing employee. It is stored searchably in the label information database 63.

The learning means 37 includes a learning model database 69, and supervised learning is performed using each set of test data acquired by the test data acquisition means 31 and each set of labels acquired by the corresponding label acquisition means 35 as teacher data. By performing supervised learning in this way, a learning model that predicts the post-employment evaluation of applicants who have taken the aptitude test is constructed. The constructed learning model is stored in the learning model database 69 in a searchable manner. For supervised learning, it is advisable to use at least one, preferably multiple, different methods. The reason why a plurality of types are preferable will be described in the description of the matching degree determining means 39 described later.

The personal history data acquisition means 38 includes at least the existing employee personal history database 65 and the existing employee interview result database 67, acquires as personal history data including the history and work history of each existing employee, and searches the existing employee personal history database 65. Store as possible. It has been experimentally proved that the history of age, gender, educational background, etc. and the work history such as the job content experienced have influenced the evaluation of the existing employees, so I hope to join the company by taking these into consideration. It is possible to provide a learning model that can predict the degree of prediction of a person's evaluation in more detail. The reason for retirement stored in the existing employee interview result database 67 is also included in the personal career. This is because the reason for retirement is closely related to individual evaluation, and the accuracy of evaluation can be improved by including this. Further, the interview process until the existing employee is hired, which is stored in the existing employee interview result database 67, may be included in the personal career. Who decided to hire and how they decided to hire are other factors that are beyond the control of existing employees, such as compatibility with interviewers and habits and tendencies of scoring, but their effectiveness is also effective. It has been proved by the experiments of the inventors.

When the learning means 39 constructs a plurality of different learning models by performing a plurality of different supervised learnings based on the same teacher data, the matching degree determining means 39 tests for each of the constructed plurality of learning models. The degree of conformity between each test label output by inputting each of the test data acquired by the data acquisition means and each evaluation label acquired by the label acquisition means is determined. By determining the degree of matching, it is possible to select and use a learning model that has the highest degree of matching, that is, is presumed to be the most conforming to the actual situation. Here is the biggest advantage of multi-supervised learning. The degree of matching is associated with each learning model and stored in the learning model database 69 in a searchable manner. When the supervised learning to be performed is singular rather than plural, the matching degree determination means 39 does not operate because it is not necessary.

(Outline configuration of forecasting means)
As shown in FIG. 1, the prediction means 5 includes an applicant test data acquisition means 51, an individual history data acquisition means, a selection means 55, an evaluation prediction means 57, a hire applicant history database 71, a hire applicant test information database 73, and a hire. It has a database of interviews with applicants and employees, and a database of evaluation prediction results after joining the company 77.

The applicant test data acquisition means 51 includes the pre-employment employee examination information database 73, and is a means for acquiring the result of the applicant for joining the company taking a predetermined job aptitude test as the applicant examination data. In the job aptitude test in the present embodiment, the trader-provided test results are acquired from the trader-provided test result database 101 outside the device and stored in the existing employee test information database 73 in the device so as to be searchable.

The applicant history data acquisition means 53 includes at least a recruitment applicant history database 71 and a recruitment applicant interview result database 75, and acquires as personal history data including the history and work history of each applicant, and the recruitment applicant history database. Stored in 71 so as to be searchable. Since it has been experimentally proved that the history of age, gender, educational background, etc. and the work history such as the job content experienced have had a great influence on the evaluation of the existing employees, these of those who wish to join the company are taken into consideration. By doing so, it is possible to provide a learning model that can predict the degree of prediction of the evaluation of prospective employees in more detail. Further, the interview process before the final interview stored in the applicant interview result database 75 may be included in the personal history. Who decided to hire and how they decided to hire are other factors that are beyond the control of existing employees, such as compatibility with interviewers and habits and tendencies of scoring, but their effectiveness is also effective. It has been proved by the experiments of the inventors.

The selection means 55 selects the learning model having the highest degree of matching from the degree of matching determined by the degree of matching means 39. A high degree of matching is presumed to be a learning model that most closely matches the actual situation of the company, so this is to be selected and used.

The evaluation prediction means 57 includes a basic information database 71 for applicants to join the company, an interview result database 75 for applicants to join the company, and a prediction result database 77 after joining the company, and the learning model when a single learning model is used, or a plurality of learning models. Is used, any of the learning models selected by the selection means 53 is used to predict the post-employment evaluation of applicants for joining the company. Basic information such as history such as name, gender, and educational background of employees before joining the company and basic information such as work history is stored in the employee basic information database 71 before joining the company, and can be obtained by accessing the database 71. The prediction result is retrievably stored in the post-employment evaluation prediction result database 77, which is the storage means.

(Execution procedure of construction method and prediction method)
This will be described with reference to FIGS. 1 and 2. Here, for the sake of clarity, this job aptitude test is referred to as the SPI (trademark) test, which is one of the tests provided by the vendor. It is not intended to exclude tests provided by other vendors or job aptitude tests developed independently.

When the program is started, the construction means 3 (test data acquisition means) of the information processing device 1 acquires "test data" of each existing employee involved in the job aptitude test (step 1). The test data is acquired by accessing the test result database 101 (Fig. 1) provided by Recruit Management Solutions Co., Ltd., which provides SPI (trademark) tests. If any of the existing employees have not taken the SPI test, it is preferable to have that person take the test to obtain the test results, but if there are circumstances such as being a retired employee and unable to contact, this is the case. It is advisable to exclude turnover from existing employees. The obtained test information is linked to each existing employee in the existing employee test information database 61 and stored in a searchable manner. In addition to the above, the input device 7 may be configured to manually input a part or all of the test data. At this time, the test data may be supplemented or modified as appropriate.

Next, the evaluation data acquisition means 33 acquires the evaluation data of each existing employee (step 3). The evaluation data is stored in the existing employee evaluation information database 62 and the existing employee interview result database 67 in a searchable manner in association with each existing employee, and is obtained by accessing this. Each evaluation data is quantified. According to the procedure described above, step 1 is first and step 3 is later, but on the contrary, step 3 may be processed first, or both steps may be processed at the same time. It does not prevent the evaluation data from being input via the input device 7 and necessary replenishment / correction.

For each of the acquired evaluation data, the label acquisition means 35 acquires the data divided into three levels (H (high), M (intermediate), and L (low) as each evaluation label (step 5). The reason for setting this level is that it was confirmed in the empirical experiments of the inventors that the accuracy is higher than the mere subjective judgment of human beings if it is divided into three in practice.

In this embodiment, "personal career data" of existing employees is also acquired (step 7). It is the personal history data acquisition means 38 that acquires the personal history data. The personal career data is the "personal career data including history / work history" defined above (the same applies to those who wish to join the company, which will be described later), and can be obtained simultaneously with or sequentially in steps 1 and 3. .. The personal history data is stored in the existing employee personal history database 65.

Next, in step 9, the type of supervised learning is determined. It is the users of the information processing device 1 who decide, but the decision is input via the input device 7. However, this decision is a tentative decision after joining the company and before making a forecast. Here, for example, it is assumed that a method of constructing a learning model with a "decision tree" that classifies and regresses using a tree structure using only "test data" is adopted.

Next, the learning means 37 creates supervised learning, that is, a prediction model (learning model) using a decision tree (step 11). Specifically, each of the test data acquired by the test data acquisition means 31 in step 1 and the three levels (H (high), M (intermediate), and L (low) acquired by the evaluation data acquisition means 33 in step 5). A prediction model is constructed by supervised learning using each set of evaluation labels divided into the above as teacher data. From the above, a learning model using the test data and evaluation data of existing employees is constructed. If the evaluation is performed after joining the company using the supervised learning model, can directly proceed to step 21, which will be described later. Here, the description will be made separately from the flowchart of FIG.

The supervised learning method using the "decision tree" adopted in step 11 is based on the examination data and evaluation data of former employees who left the job together with the currently working employees, so a highly accurate learning model can be used. There is something to offer, but it is not without room for greater accuracy without compromising the benefits of supervised learning techniques. It is to perform machine learning of a method different from the decision tree using only the test data in the above example under the same conditions, compare the results, and select the best one. This point will be described in the following sections.

Return to the flowchart of FIG. The evaluation prediction means 55 of the prediction device 5 predicts the evaluation after joining the company using the prediction model constructed earlier (step 13). The selection means 53 is made to determine whether the accuracy or accuracy of the prediction result is at a predetermined level (step 15). If it is determined that the prediction accuracy is not sufficient, including the meaning just in case, return to step 9 and use a procedure other than the "decision tree", for example, "random forest", in the same procedure as the "decision tree". To determine the prediction accuracy (step 17). If it is determined that the accuracy is sufficient at the "random forest", the process proceeds to step 19 and the "random forest" in this case is selected as the "optimal prediction model" (learning model). In this case, different supervised learning (eg, neural networks, deep learning, etc.) is performed based on the assumption that even if the accuracy exceeds a certain level, there may be a higher accuracy. Different learning models may be constructed and the judgment of step 15 may be made for each of these learning models.

The reason why multiple learning models are prepared and verified as described above is that algorithms such as "decision tree" and "random forest" have different accuracy depending on the type and number of data used. Because I know. In order to compensate for the decrease in accuracy due to this difference, a more accurate learning model is searched for for each algorithm by using a method in which the type and number of data used are changed. According to this method, by verifying each of the learning models by different supervised learning, it is possible to select the optimum prediction model that is more in line with the actual situation of the company, and by extension, it is possible to make highly accurate post-employment prediction. Connect. It should be noted that the optimum prediction model can be appropriately supplemented / modified by the user via the input device 7. For example, the ability of leadership to make corrections in consideration of the downsides such as the fact that the results were not achieved due to long-term medical treatment due to illness or that the compatibility with a specific boss was very bad, and the trust of colleagues is strong. It is also possible to make additions based on the positive side such as there is.

By the way, when the optimum prediction model is selected in step 19, the applicant test data acquisition means 51 accesses the pre-employee employee examination information database and acquires the examination data of the applicant for joining the company (step 21). Step 21 may be executed between steps 17 and 19, or may be executed at the same time.

Next, the applicant history data acquisition means 53 accesses the applicant history database 71 and the applicant interview result database 75 to obtain data related to the history of applicants, work history, interview process leading up to the final interview, and the like. Acquire (step 23). The data acquisition in step 23 may be performed before or before steps 19 and 21, and can be executed at the same time. Experimentally, the history of age, gender, educational background, and work experience (in the case of new graduates, internships, volunteer activities, etc.) had a great influence on the evaluation of existing employees. It has been proved as mentioned above, and this also applies to applicants for joining the company. By importing such historical data, it becomes possible to make detailed evaluations such as evaluations by department, which would not have been possible without it.

When the necessary data and information are prepared as described above, the evaluation prediction means 55 predicts the post-employment evaluation of the applicant for joining the company based on each of the applicant test data of the applicant for joining the company and the optimum prediction model (learning model). (Step 25). The prediction result is displayed on the display means 9 (step 27). This ends the program. By being displayed on the display means, the user (expected person) who sees this will know the expected evaluation after joining the company, and along with this, will get a powerful clue for selecting the employer. be able to.

The above-mentioned optimal forecast model is for post-employment evaluation in a company to which existing employees belong, but for post-employment evaluation of other companies that have the same or similar corporate environment such as industry, business type, and scale. May be used for. From a different standpoint, it is said that the learning model constructed by the above method can be used for post-employment evaluation of one's own company in another company that has the same or similar corporate environment such as industry, business type, and scale. It will be. In addition, on the condition that the type of business and business type are at least the same or similar, all of the multiple companies with a relatively small number of employees are integrated and regarded as one integrated company, and so far among the integrated companies. It will be possible to build the learning model described above and predict the evaluation after joining the company. For this point, refer to this modification described later.

As defined above, since this specification considers "applicants to join the company" in a broad concept, it also includes those who have not yet been ordered or requested to transfer to another department by existing employees. Therefore, it is added that the learning model and post-employment evaluation constructed by the above procedure can also be used to predict post-department evaluation for those who are moving within the company and are going to change departments due to transfer etc. I will do it.

(Modified example of this embodiment)
A modified example of the present embodiment (hereinafter, referred to as “the present modified example”) will be described with reference to FIG. The evaluation prediction device 1'(hereinafter referred to as" prediction device 1'" shown in FIG. 2 is different from the prediction device 1 shown in FIG. 1 and the basic structure 1'in that the other company's existing employee database 103 owned by the prediction device 1'is used. The learning model construction means 3'(hereinafter referred to as" construction means 3'" of the prediction device 1') handles the points that the prediction device 1 does not have and the processing of the data obtained from the existing employee database 103 of another company that the prediction device 1 did not perform. There are two points to do. Except for these two points, the evaluation device 1 and the evaluation committee device 1A are common. Therefore, for this common part, the reference numerals used in FIG. 1 are also used in FIG. 2, and these explanations will be omitted to the extent possible.

The construction means 3'compared with the construction means 3 has a function of processing the in-house data of the adopting company and the test data of another company, which will be described later. Therefore, the reference numerals and names shown in FIG. 1 are shown in FIG. 2 as they are. That is, the construction means 3'contains with the test data acquisition means (other company's test data acquisition means) 31, the evaluation data acquisition means (other company's evaluation data acquisition means) 33, the label acquisition means (other company's label acquisition means) 35, and the learning means 37. The degree determination means 39 is provided, and as a preferred embodiment, the personal history data acquisition means 38 is further provided.

In FIG. 2, the existing employee database 103 of other companies is defined by each of the current or past employees of one or more other companies (hereinafter referred to as “existing employees of other companies”) different from the hiring companies targeted in this embodiment. The result of taking the job aptitude test is stored as the data of other companies, and is communicably connected to the prediction device 1'via the network N. The test data acquisition means 31 acquires the result of the job suitability test (test data of another company) received by an existing employee of another company from the existing employee database 103 of another company and stores it in the existing employee test information database 61 in a searchable manner. The evaluation data acquisition means 33 acquires other company's evaluation data indicating the result of internal evaluation of each other company's existing employee from the other company's existing employee database 103, and stores it in the existing employee evaluation information database in a searchable manner.

Here, together with FIG. 2, FIG. 4 related to this modification will be described. The difference between the flowchart of FIG. 4 and the flowchart of FIG. 3 is that it has three steps, that is, step 9', step 10 and step 10', which are not included in the present embodiment (FIG. 3) and acquire data of other companies. Since there are only points, I will explain these. In step 9'provided after step 9, it is confirmed whether the test data acquisition means 31 has a sufficient amount (number of people) of data in the hiring company to build a learning model for evaluating applicants for joining the company. To do. A sufficient amount of data changes depending on the type of business, business type, company size, social situation, etc. and is left to the judgment of the user, but for example, if one organization has less than 30 people, it is not enough. There will be. If the amount of data is sufficient, the process proceeds to step 11 as it is.

On the other hand, if the amount of data is insufficient in step 10, the existing employee test information means 31 accesses the database 61 to acquire the test data of another company, and creates data combined with the test data of the company's employee (step). 10'). At the same time, the evaluation data acquisition means 33 accesses the existing employee evaluation information database 62 to acquire the evaluation data of other companies (step 10), and creates data combined with the evaluation data of the company's employees (step 10'). Then proceed to step 11. The steps before and after the above three steps are the same as the processes in the present embodiment. Here, if the company's data does not exist at all, all the data of other companies will be used. It is important to pay sufficient attention to the security of confidential information such as personal information, especially when handling data from other companies.

<Example 1>
With reference to FIG. 5, a demonstration example performed using the construction method / evaluation prediction method according to the present invention is shown. First, in the first embodiment, a demonstration was conducted targeting a Web service company, a real estate industry, an IT-related company, a human resources industry, and a construction industry. The number of existing employees, the job aptitude test used, and the results are shown in Fig. 5. Looking at the correct answer (match) rate and prediction accuracy, the correct answer rate is 59.3% for Web service companies, the prediction accuracy (Kappa coefficient, the same applies hereinafter) 0.24, and the correct answer rate is 60.0% for the real estate industry, and the prediction accuracy is 0. .31. Correct answer rate 62.4% / prediction accuracy 0.41 for IT-related companies, correct answer rate 51.0% / prediction accuracy 0.26 for human resources industry, and correct answer rate 68.9% / prediction for construction industry The accuracy was 0.45.

<Present Example 2>
On the other hand, FIG. 6 shows the retirement prediction accuracy rate and prediction accuracy of the real estate industry and the IT-related industry. The number of existing employees, the job suitability test used, and the results are shown in Fig. 6. Correct answer (match) rate and prediction accuracy in retirement prediction are 91.0% correct answer rate for real estate industry, prediction accuracy (Kappa coefficient, the same applies hereinafter) 0.33, recall rate 61.5%, and IT related industry (about 1000 people) ), Correct answer rate 87.7%, prediction accuracy 0.14, recall rate 49.1%, and also in IT-related industry (about 300 people), correct answer rate 89.7%, prediction accuracy 0.30, recall rate 72.7. %Met.

<Evaluation of Example 1>
In this embodiment, the labels used are three levels of H (high), M (intermediate), and L (low). Therefore, in the case of prediction without using the learning model, the accuracy rate is about 33% (one-third). ). In each empirical example, the match rate is much higher than about 33%, and it can be confirmed that the evaluation prediction using the learning model is effective. In addition, although there are differences in the company size, industry, and job aptitude test used in each empirical example, it has been shown that evaluation prediction is effective by using this method in the same way. In addition, by using the data of "affiliation" and "job type" in addition to the data of each job aptitude test, the matching rate and prediction accuracy (Kappa coefficient, explained in the following section) are further improved. Although it has been shown that job aptitude test data alone can provide sufficiently useful results, it is more accurate by adding "personal background data", constructing multiple learning models, and selecting the optimal prediction model. It was confirmed that a high learning model can be constructed, and it was found that this method is sufficiently practical.

I will add about the Kappa coefficient. The Kappa coefficient is an index originally created to confirm "whether or not the judgments match when two parties observe one object", and the closer the value is to 1, the higher the degree of agreement. Means that. A well-known paper on this Kappa coefficient, "Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977; 33 (1): 159-174.", 0.0-0.2: Slight agreement. , 0.21 to 0.40 fair agreement, 0.41 to 0.60 moderate agreement, 0.61 to 0.80 substantial agreement, and 0.81 to 1.0 almost perfect or perfect agreement ) Is shown.

Looking back at FIG. 5 and considering the Kappa coefficient shown in this demonstration example, even the smallest Web service company shows 0.24, and the real estate industry shows 0.31, both showing "reasonable agreement". Furthermore, a certain degree of prediction accuracy is shown. Furthermore, for IT-related companies, 0.41 has reached the standard of "moderate agreement". Of course, since this demonstration example is a future prediction rather than a past verification, it cannot be an "exact match", and it is considered that a "tendency (= more than a reasonable match)" is a sufficient result. Be done. Therefore, it was reconfirmed that this method is sufficiently practical.

<Evaluation of Example 2>
In the first embodiment, the in-house evaluation was divided into three levels, H (high), M (intermediate), and L (low), and the evaluation level was defined. On the other hand, the retirement related to the second embodiment shown in FIG. In the empirical example of predicting, the label used was a two-level level of "retirement" and "employment". In other words, it is the result of supervised learning using the data test data of other companies for each employee of the company and the pair of "retirement" and "employment" of each employee as teacher data.

The percentage of "retirement" varies depending on the type of industry and employment type, but it is often around 10%. Therefore, in the case of a model that does not use a learning model, the recall rate (explained in the next section and below) is around 10%. The recall rate in each empirical example is as large as 10%, and it can be confirmed that the evaluation prediction using the learning model is effective. In addition, as in the above-mentioned demonstration example, this method is effective even if the company size, industry, and job aptitude test used are different, and by using data of "grade" and "job type" in addition to each job aptitude test data. , The correct answer rate, Kappa coefficient, and recall rate have been shown to increase.

I would like to add to the "recall rate" of retirement forecasts. The “recall rate” is a value indicating the percentage of those who actually “retired” among those who predicted “retirement”. It can be considered that the higher the recall rate, the more the person who "retirs" is identified. For example, if the "recovery rate" is high, the "percentage of people who predicted retirement but do not actually retire" is low, so the probability of taking measures to prevent "retirement" for "people who do not retire" is low. Therefore, it becomes possible to efficiently implement measures for "only those who retire".

Considering the recall rate in the retirement forecast for Example 2 (see the rightmost column of the chart in Fig. 6), the largest one is 72.7% of the IT-related companies (300 sales employees only) in the bottom column of the chart. Next is 61.5% of the real estate industry (about 300 new graduates only) at the top of the chart. Even the smallest recall rate is 49.1% of the other IT-related industry (about 1000 employees) shown in the chart interruption. From the above empirical results, it can be seen that about 1 in 2 people who predicted "retirement" actually retired. The actual retirement rate of the target company this time is also around 10%, and the recall rate of 49.1% means that the retirement rate can be predicted with a probability more than four times the actual retirement rate. This shows that the efficiency of taking measures to prevent actual "retirement" is more than four times higher, which shows that this method is practical.

The first and second embodiments described above are for the present embodiment in which only the company's data is targeted. On the other hand, when targeting merged data obtained by merging other companies'data with own data, or data consisting only of other companies' data (merged data with other companies' data when the company's data is set to zero), the industry / business type of the other company If etc. are similar to those of the adopting company, it is presumed that the same or similar result as that of Examples 1 and 2 above can be obtained.

1 Evaluation prediction device 1'Evaluation prediction device 3 Learning model construction device (construction device)
3'Learning model construction device (construction device) related to this modification
5 Post-employment evaluation prediction device (prediction device)
31 Test data acquisition means (other companies' test data acquisition means)
33 Evaluation data acquisition means (evaluation data acquisition means of other companies)
35 Label acquisition means 37 Learning means 38 Personal background data acquisition means 39 Matching degree judgment means 51 Applicant examination data acquisition means 53 Selection means 55 Evaluation prediction means 57 Interview data acquisition means 61 Existing employee examination information database 62 Existing employee evaluation information database 63 Label information database 65 Existing employee personal history database 67 Existing employee interview result database 69 Learning model database 71 Pre-employment basic information database 73 Pre-employment examination information database 75 Pre-employment employee interview database 77 Post-employment evaluation prediction result database 101 Contractor-provided test result database 103 Existing employee database of other companies N network

Claims (13)

Test data to acquire the results of each current or past employee (hereinafter referred to as "existing employee") of the company scheduled to be hired (hereinafter referred to as "hiring company") to take the prescribed job aptitude test as test data. Acquisition method and
Evaluation data acquisition means to acquire evaluation data showing the results of internal evaluation of each existing employee,
The evaluation data acquired by the evaluation data acquisition means is divided into three levels of H (high), M (intermediate), and L (low), and the label acquisition means for acquiring each evaluation label and the test. By supervised learning using each set of test data acquired by the data acquisition means and each set of labels acquired by the corresponding label acquisition means as supervised data, the appropriateness is desired to join the hiring company. It is equipped with a learning means to build a learning model for post-employment evaluation prediction of applicants who have taken the exam.
The evaluation data is a learning model construction device characterized in that the evaluation data is data in which the turnover of the existing employee is negatively evaluated. Further equipped with a means for acquiring personal history data to be acquired as personal history data including the history and work history of each of the existing employees.
Claim 1 is characterized in that the learning means is configured to perform supervised learning using the group to which the personal history data of each of the existing employees acquired by the personal history data acquisition means is added as teacher data. The described learning model construction device. The learning means is configured to construct a plurality of different learning models by performing a plurality of different supervised learning using the teacher data as teacher data.
Besides,
The degree of matching between each test label output by inputting each test data acquired by the test data acquisition means into each of the plurality of learning models constructed by the learning means and each evaluation label acquired by the label acquisition means is determined. A matching degree determination means for determination is provided.
The learning model building apparatus according to claim 1 or 2, wherein the learning model is constructed. Other companies'examinations in which current or past employees of one or more other companies (hereinafter referred to as "existing employees of other companies"), which are different from the above-mentioned hiring companies, acquire the results of taking the prescribed job aptitude examination as other company's examination data. Data acquisition method and
Further equipped with another company's evaluation data acquisition means to acquire other company's evaluation data showing the results of internal evaluation of each existing employee of the other company.
The label acquisition means is obtained by dividing each of the merged data, which is a combination of the evaluation data (in-house evaluation data) acquired by the evaluation data acquisition means and the evaluation data of another company acquired by the evaluation data acquisition means of the other company, into a plurality of levels. , The learning model construction apparatus according to any one of claims 1 to 3, wherein the data is acquired as each evaluation label. Other company's test data that acquires the results of each current or past employee (hereinafter referred to as "existing employee of another company") of one or more other companies, which is different from the hiring company, taking the prescribed job aptitude test as other company's test data Acquisition method and
Other company's evaluation data acquisition means to acquire other company's evaluation data showing the results of other company's internal evaluation of each existing employee of the other company,
Each of the other company's evaluation data acquired by the other company's evaluation data acquisition means is divided into three levels of H (high), M (intermediate), and L (low), and the other company's label acquisition means is acquired as each evaluation label. ,
Joined the hiring company by supervised learning using each set of other company's test data acquired by the other company's test data acquisition means and each set of labels acquired by the other company's label acquisition means corresponding to these as other company's teacher data. A learning method of other companies that builds a learning model for post-employment evaluation prediction of applicants who have taken the above-mentioned aptitude test in hopes of
With
The evaluation data is data that negatively evaluates the turnover of the existing employee.
A learning model construction device characterized by the fact that A post-employment evaluation prediction device using a learning model constructed by the learning model construction device according to any one of claims 1, 2, 4, and 5.
Applicant test data acquisition means for pre-hiring employees to acquire the results of taking the prescribed job aptitude test as applicant test data,
It is characterized by being provided with an evaluation prediction means for predicting the post-employment evaluation of the applicant for joining the company based on each of the applicant examination data of the pre-employment employee acquired by the applicant examination data acquisition means and the learning model. Post-employment evaluation prediction device. A post-employment evaluation prediction device using a plurality of learning models constructed by the learning model construction device according to claim 3.
Applicant test data acquisition means for acquiring the results of the applicants taking the prescribed job aptitude test as applicant test data, and
A selection means for selecting the learning model with the highest degree of matching from the degree of matching determined by the matching degree determining means, and
It is characterized by being provided with an evaluation prediction means for predicting the post-employment evaluation of the applicant for joining the company based on each of the applicant examination data of the applicant for joining the company acquired by the applicant examination data acquisition means and the learning model. Post-employment evaluation prediction device. It is a post-employment evaluation prediction method executed by a computer.
Test data to acquire the results of each current or past employee (hereinafter referred to as "existing employee") of the company scheduled to be hired (hereinafter referred to as "hiring company") to take the prescribed job aptitude test as test data. Acquisition step and
Evaluation data acquisition step to acquire evaluation data showing the result of internal evaluation of each existing employee,
A label acquisition step in which each of the evaluation data acquired in the evaluation data acquisition step is divided into three levels of H (high), M (intermediate), and L (low) is acquired as each evaluation label.
We hope to join the hiring company by supervised learning using each set of test data acquired in the test data acquisition step and each set of labels acquired in the corresponding label acquisition step as teacher data. Learning steps to build a learning model for post-employment evaluation prediction of applicants who have taken the aptitude test, and
With
The evaluation data is data that negatively evaluates the turnover of the existing employee.
A learning model construction method characterized by this. Further equipped with a personal history data acquisition step to be acquired as personal history data including the history and work history of each existing employee.
The learning model construction method according to claim 8, wherein in the learning step, supervised learning is performed using the group in which the personal history data of each of the existing employees acquired in the personal history data acquisition step is added as teacher data. .. The other company's test data acquisition step to acquire the results of each existing employee of another company taking the prescribed job aptitude test as other company's test data,
Further equipped with another company's evaluation data acquisition step to acquire other company's evaluation data showing the results of internal evaluation of each existing employee of the other company.
In the label acquisition step, each evaluation label is obtained by dividing each of the merged data, which is a combination of the evaluation data acquired in the evaluation data acquisition step and the evaluation data of another company acquired in the evaluation data acquisition step of the other company, into a plurality of levels. The learning model construction method according to any one of claims 8 or 9, wherein the data is acquired as. It is a post-employment evaluation prediction method executed by a computer.
The results of each current or past employee of one or more other companies (hereinafter referred to as "existing employees of other companies"), which is different from the company scheduled to be hired, have taken the prescribed job aptitude test as other company test data. Other companies' test data acquisition steps to be acquired and
Other company's evaluation data acquisition step to acquire other company's evaluation data showing the result of each other company's internal evaluation of each existing employee of the other company,
The other company's label acquisition step, in which each of the other company's evaluation data acquired in the other company's evaluation data acquisition step is divided into three levels of H (high), M (intermediate), and L (low), is acquired as each evaluation label. ,
Joined the hiring company by supervised learning using each of the other company's test data acquired in the other company's test data acquisition step and the corresponding set of labels acquired in the other company's label acquisition step as supervised learning. Other companies' learning steps to build a learning model for post-employment evaluation prediction of applicants who took the above-mentioned aptitude test in hopes of
With
The evaluation data is data that negatively evaluates the turnover of the existing employee.
A learning model construction method characterized by the fact that. In the learning step, a plurality of different learning models are constructed by performing a plurality of different supervised learning using the teacher data as teacher data.
Besides,
The degree of matching between each of the test labels output by inputting each of the test data acquired in the test data acquisition step into each of the plurality of learning models constructed in the learning step and each of the evaluation labels acquired in the label acquisition step is determined. A matching degree determination step for determination is provided.
The learning model construction method according to any one of claims 8 or 9, wherein the learning model is constructed. A post-employment evaluation prediction method using a plurality of learning models constructed by the learning model construction method according to claim 12.
Applicant test data acquisition step to acquire the result of the applicant who wants to join the company taking the prescribed job aptitude test as applicant test data, and
A selection step for selecting the learning model with the highest degree of matching from the degree of matching determined in the matching degree determination step, and
It is characterized by including an evaluation prediction step for predicting the post-employment evaluation of the applicant for joining the company based on each of the applicant examination data of the applicant for joining the company acquired in the data acquisition step for the applicant for joining the company and the learning model. Post-employment evaluation prediction method.

Images