JP7368902B1 - Human resources evaluation system - Google Patents

Abstract

[Problem] We propose a human resources evaluation system that can quantitatively perform evaluations that are directly linked to organizational results. [Solution] Answer sentences input by training participants to specific questions are converted into sentence vectors, and the cosine similarity between the specific participant's sentence vector and each participant's sentence vector is calculated, and each It is designed to obtain evaluation scores for evaluating participants. According to this configuration, other participants can be quantitatively evaluated based on a specific participant who contributes to the results of the organization, so that evaluations that are directly linked to the results can be performed. [Selection diagram] Figure 4

Description

The present invention relates to a human resources evaluation system for evaluating participants who have participated in training.

In organizations such as companies, human resource development and education are important, and there is an increasing need for human resource evaluation to support these efforts. Conventionally, such human resource evaluations were carried out by personnel personnel, etc., and therefore tended to be intuitive evaluations by the person in charge, making it difficult to perform quantitative evaluations.

On the other hand, in recent years, with the rapid development of computers and analysis software, evaluation systems that use these tools to evaluate human resources have been proposed. For example, in the configuration of Patent Document 1, a computer analyzes reports created by each employee who participated in in-house training using natural language processing, and determines the relative degree of relevance for each sentence of multiple items included in each report. Then, each employee's performance is estimated based on this degree of association. Here, the report includes each sentence input for a plurality of items, and each sentence is analyzed by the natural language processing. Based on the results of analyzing each sentence, the relative degree of relevance of each sentence is determined, and a plurality of performances are estimated based on the degree of relevance. According to this configuration, the plurality of performances can be estimated for each employee, which can be useful for human resource development and education.

JP2022-64238A

In the conventional configuration of Patent Document 1 mentioned above, multiple performances are estimated based on reports created by employees (the degree of relevance of each sentence to multiple items), so detailed support suitable for each employee is provided. Although it has the advantage of being easy to perform, it is difficult to evaluate these multiple performances that can be directly linked to the results desired by the organization. This is because the performance is estimated for each employee based on the relevance of each sentence entered for multiple items, so the performance is estimated based on the employee's personal abilities (e.g. planning ability, persistence, This is because it cannot clearly be said that there is a direct connection with the above results because it indicates a positive attitude (e.g., proactiveness). In this way, with conventional evaluations, it is difficult to evaluate human resources that can directly contribute to the growth and development of an organization.

The present invention proposes a human resources evaluation system that can quantitatively perform evaluations that are directly linked to the above-mentioned results.

The present invention is a human resource evaluation system that evaluates a plurality of participants who have participated in a predetermined training, and the system evaluates a plurality of participants who have participated in a predetermined training. an answer storage means that stores the answers in association with the participants; and a proportion of the answer sentences of each participant that are related to the answer sentences of a specific participant selected in advance from among the participants who have participated in the training. an evaluation score acquisition means for obtaining an evaluation score for evaluating each participant based on the training, and the evaluation score acquisition means analyzes the answer sentences of each participant who participated in the training by natural language processing, a vectorization processing means for obtaining a text vector of each response text by converting it into text vectors based on the analysis; and a cosine of the text vector of each participant obtained by the vectorization processing means and the text vector of the specific participant. This human resources evaluation system is characterized by comprising: similarity calculation means for calculating the respective degrees of similarity and obtaining the evaluation score based on the cosine similarity of each participant.

In such a configuration, an evaluation score is obtained by calculating the cosine similarity between a participant's sentence vector and a specific participant's sentence vector. Evaluate people. Thereby, by appropriately setting specific questions, it is possible to quantitatively learn the differences in behavioral characteristics (behavior patterns, thinking tendencies, etc.) between a specific participant and each participant. Then, if a specific participant is someone who contributes to the organization's results (such as a person with excellent sales performance or a developer of an excellent product), the evaluation score can be used to quantify the evaluation that is directly linked to the result. It can be carried out. In particular, since the present invention is based on specific participants who participated in the same training, accurate evaluation can be performed. Here, the training only needs to have at least the same training content.

Furthermore, according to the configuration of the present invention, it is also possible to evaluate behavioral changes (changes in behavioral characteristics) of each participant by obtaining evaluation scores for multiple training sessions and looking at chronological changes. . Furthermore, by comparing evaluation scores for multiple training sessions, it is also possible to evaluate the usefulness of each training session and the instructor.

In the human resource evaluation system of the present invention described above, at least two specific questions are set as the specific questions, and the vectorization processing means converts the answer text of each specific question into a text vector for each participant. and a processing means for obtaining a composite sentence vector for each participant by composing these sentence vectors, and the similarity calculation means is configured to combine the composite sentence vector for each participant and the specific participant. A configuration is proposed that includes processing means for calculating a cosine similarity with the synthetic sentence vector and obtaining the evaluation score of each participant based on the cosine similarity.

In such a configuration, the cosine similarity between each participant and a specific participant is calculated using a composite sentence vector obtained by synthesizing sentence vectors obtained by converting multiple answer sentences, so that one question can be answered from a plurality of specific questions. An evaluation score is obtained. Since evaluation can be performed using this one evaluation score, evaluation can be made more clearly and easily.

As described above, the human resource evaluation system of the present invention evaluates each participant based on a specific participant, and therefore it is possible to quantitatively obtain the difference in behavioral characteristics between each participant and the specific participant. can. Thereby, it is possible to determine participants who have the same behavioral characteristics as a specific participant who contributes to the results of the organization, and therefore it is possible to perform evaluations that are directly linked to the results.

1 is a schematic diagram showing the configuration of a human resource evaluation system 1 according to the present embodiment. 2 is a block diagram showing the functions of the evaluation management device 2 of the human resource evaluation system 1. FIG. 2 is an explanatory diagram showing answer text data stored in the database 12 of the evaluation management device 2. FIG. FIG. 3 is a flow diagram showing evaluation score processing for obtaining evaluation scores of participants. It is a chart showing evaluation scores of each participant. It is an explanatory diagram showing a graph of evaluation scores of participants.

Embodiments embodying the present invention will be described with reference to the accompanying drawings.
FIG. 1 shows a human resources evaluation system 1 of this embodiment. The human resource evaluation system 1 of this embodiment is for evaluating employees of a company who have participated in training, and is based on a communication network (such as the Internet) 8 that is used by participant terminals 7 used by participants who have participated in training. The evaluation management device 2 is equipped with a communication-capable evaluation management device 2. The participant terminal 7 is a personal computer, a tablet, a smartphone, or the like having a communication function, and has application software installed thereon that allows the user to input answer texts, which will be described later.

The evaluation management device 2 includes a management server 11, a database 12, and a transmitting/receiving means 13. The management server 11 is composed of one or more computers having a central control unit (CPU), a storage device (ROM, RAM), and the like. The database 12 is constituted by a storage device that stores and holds programs executed by the management server 11, data used by the programs, data generated by the programs, and the like. The transmitting/receiving means 13 has a function of transmitting and receiving data and signals to and from the participant terminals 7 via the communication network 8 .

As shown in FIG. 2, the database 12 includes an answer storage unit 15 that stores and holds the data of answer texts input via the communication network 8, and a vector data storage that stores and holds text vectors obtained by vectorizing the answer texts . 16, and an evaluation score storage section 17 that stores and holds evaluation scores, which will be described later. Here, in each of the storage units 15 to 17, data of each answer text is stored in association with each employee. In this embodiment, as will be described later, each employee inputs answers to two specific questions, so the answer storage section 15 stores the answers as shown in FIG. , data of two answer sentences are stored for each employee.

As shown in the functional block diagram of FIG. 2, the management server 11 includes an evaluation score acquisition means 21 having a vectorization processing means 22 and a similarity calculation means 23, and also receives evaluation scores obtained by the evaluation score acquisition means. It is provided with an evaluation display means 24 for displaying the evaluation. This evaluation score acquisition means 21 is a processing function executed by the program and data read from the database 12 by the central control unit (CPU) described above. An evaluation score is calculated from the data of the answer text .

The vectorization processing means 22 analyzes the answer text by natural language processing, and converts it into a text vector based on the analysis result. This text vector is then stored in the vector data storage section 16 of the database 12. To be more specific, the answer sentence data is read from the answer storage section 15 of the database 12 and analyzed by natural language processing. For natural language processing, general analysis processing such as morphological analysis and syntactic analysis can be applied. Then, using the analysis results obtained through this natural language processing, processing is performed to convert the answer text into a text vector. In this embodiment, Sentence-Bert is used for such natural language processing and text vectorization processing. This converts the answer text into a text vector. In this way, the vectorization processing means 22 converts the answer sentences of each employee who participated in the training into sentence vectors. As mentioned above, in this embodiment, each employee stores two answer text data, so the vectorization processing means 22 converts the two answer texts into text vectors for each employee. Obtain two sentence vectors.

The similarity calculation means 23 synthesizes two sentence vectors for each employee to obtain one composite sentence vector. Then, using the predetermined composite sentence vector of a specific employee as a reference, the degree of cosine similarity between the composite sentence vector of the specific employee and the composite sentence vector of each other employee is calculated, and this calculated value is set to 0. Obtain an evaluation score (hereinafter referred to as an overall evaluation score) standardized within the range of ~1 (0 being the same as the specific employee and 1 being the most different). Then, the overall evaluation score of each employee is stored in the evaluation score storage section 17 of the database 12. Here, in this embodiment, at least one excellent employee (hereinafter referred to as a high performer) in the company is determined as the specific employee from among the employees who participated in the training. For example, employees with excellent sales performance or employees who plan and develop excellent products are selected as high performers.

Furthermore, the similarity calculation means 23 of this embodiment calculates the cosine similarity between each employee and the hyper-performer for one of the two text vectors, and similarly calculates the cosine similarity between each employee and the hyper-performer for the other text vector. Calculate the cosine similarity of each. Then, for each employee, one cosine similarity and the other cosine similarity are standardized so that the maximum value is 1 (hereinafter referred to as standardization), and two evaluation scores (first evaluation described later) are standardized. score and a second evaluation score). These evaluation scores are also stored in the evaluation score storage 17 of the database 12 in the same way as the comprehensive evaluation score described above.

Note that the cosine similarity calculation in the similarity calculation means 23 can use a conventionally known method, so the details will be omitted.

The evaluation display means 24 is a processing function that displays evaluation results based on data read from the evaluation score storage section 17 of the database 12 on a display means (not shown) (such as a monitor or a terminal for an administrator). Specifically, the evaluation scores are displayed in a table (FIG. 5) or a graph (FIG. 6).

Next, the flow of human resources evaluation using the human resources evaluation system 1 of this embodiment will be explained using the following specific example.
For example, when evaluating the human resources of employees who have participated in a training program that is held a specified number of times for employees of a company, two specific questions predetermined for each session are asked, and each The answer text is input from each participant terminal 7. Here, it is preferable that the two specific questions have a relationship in which one is related to the other, and more preferably, they are non-factoid type questions. The content of the question may be the same for all sessions, or the content of the question may be different for each session. In this example, the first specific question is ``What did you notice this time?'' and the second specific question is ``Why did you notice that?'' Ask specific questions and have them enter answers. Here, the answer text to the first specific question is called the first answer text, and the answer text to the second specific question is called the second answer text.

Furthermore, at least one high performer is selected from among the employees participating in the training. For example, if two employees a and b are selected, these employees a and b are set as high performers and stored in the database 12.

After each training session, each employee who participated in the training is asked to input the first and second answer sentences. As shown in FIG. 3, the response storage section 15 of the database 12 stores and stores two response texts input from each employee in association with each employee. In FIG. 3, only the data of the first and second answer sentences for one training session are described, but after the training is completed, similar answer sentences are stored for each session.

When response sentences are input from each employee at the end of each training session or after the end of training, the management server 11 executes evaluation score processing (FIG. 4) to calculate the evaluation score of each employee.

In the evaluation score processing, the vectorization processing means 22 reads the data of the first and second answer sentences for each employee from the answer storage section 15 of the database 12 (S10), and the data of the first and second answer sentences are read in for each employee. are converted into text vectors (S20). Then, the first text vector obtained by converting the first answer text and the second text vector obtained by converting the second answer text are stored in the vector data storage section 16 of the database 12 in association with each employee. As a result, the vector data storage unit 16 stores the first and second sentence vectors for each employee. Here, text vectorization processing is performed on answer texts input from all employees, including the hyper-performers.

Thereafter, the above-described similarity calculation means 23 reads the first and second sentence vectors of each hyperformer a and b from the vector data storage unit 16, and a composite sentence is obtained by combining the first sentence vector and the second sentence vector. A vector is generated (S40). Similarly, for each other employee, the first and second sentence vectors are read to generate a composite sentence vector (S40), and for each employee, the composite sentence vector of the employee and the composite sentence vector of the hyperformers a and b are generated. The cosine similarity is calculated to obtain the above-mentioned comprehensive evaluation score (S50). As a result, a comprehensive evaluation score for each employee based on the high performers a and b is obtained. The overall evaluation score of each employee is stored in the evaluation score storage section 17 of the database 12 in association with each employee.

In this way, one evaluation score is obtained for each employee each time the training is conducted. This evaluation score is displayed by the above-mentioned evaluation display means 24, for example, in the chart shown in FIG. Here, the evaluation score (numeric value) closer to "0" indicates that it has the same behavioral characteristics as hyper performers a and b. The total evaluation score is the average value of each implementation. By looking at these evaluation scores, it is possible to perform a quantitative human resource evaluation for each employee. In addition, by looking at the evaluation scores for each implementation session, you can check the growth (behavioral change) of each employee. Specifically, since the evaluation score of employee o is gradually decreasing, it can be confirmed that employee o has made great progress through the training. In this embodiment, an evaluation score is not calculated based on the content of the first training session (no specific questions are asked).

Further, the similarity calculation means 23 calculates, for each employee, the cosine similarity between the first sentence vector of the employee and the first sentence vector of the hyper performers a and b, and also calculates the cosine similarity between the first sentence vector of the employee and the first sentence vector of The cosine similarity between the vector and the second sentence vector of hyperperformers a and b is calculated. Then, by standardizing the cosine similarity of the first sentence vector and the cosine similarity of the second sentence vector, the first evaluation score that has standardized the cosine similarity of the first sentence vector and the cosine similarity of the second sentence vector A second evaluation score in which the degree is standardized is obtained (S50). As a result, the first and second evaluation scores for each employee are obtained, and these are stored in the evaluation score storage section 17 in association with each employee.

These first and second evaluation scores can be displayed, for example, in the graph shown in FIG. 6 by the evaluation display means 24 described above. Here, the black dots on the graph are the data of hyper performers a and b. This graph is a graph of the standardized first and second evaluation scores as described above, and a positive correlation can be seen between the first and second evaluation scores. This allows you to visually understand the differences in behavioral characteristics between High Performers A and B and other employees, and by setting appropriate thresholds on the vertical and horizontal axes, you can identify employees who are expected to grow. Can be easily identified.

In this example, a chart (Figure 5) that displays the evaluation score of each employee and a graph (Figure 6) that displays the first and second sentence vectors are illustrated. It is also possible to evaluate on a scale of 5 (for example, 5 levels), or to calculate and display the growth rate from the evaluation score for each implementation. Furthermore, if training for employees is conducted with different content or timing, it is also possible to evaluate the training content and instructors by checking the trends in evaluation scores for multiple training sessions.

In the human resource evaluation system 1 of this embodiment, as mentioned above, the evaluation score of each employee is calculated based on the high performers selected from among the employees who participated in the training, so the evaluation score The difference in behavioral characteristics between each employee and the high performer can be expressed quantitatively. This makes it possible to determine employees who have behavioral characteristics close to those of high performers. Here, high performers are excellent employees at a company, such as employees with excellent sales results or employees who plan and develop excellent products, so employees with behavioral characteristics similar to these are expected to help companies grow. It can be evaluated as contributing to development and development. In addition, since the high performers are selected from among employees who have participated in the same training, it is possible to evaluate employees accurately, and the first and second specific questions are suitable for investigating the behavioral characteristics mentioned above. Since it is set to , it is possible to appropriately obtain the difference in behavioral characteristics from the high performer. Therefore, according to the configuration of this embodiment, it is possible to quantitatively perform an evaluation that is directly linked to the results desired by a company.

In the above embodiment, the employees of the company who participated in the training correspond to the participants according to the present invention, and the high performers (specific employees) correspond to the specific participants according to the present invention. The answer storage section of the database 12 corresponds to answer storage means according to the present invention. Further, the first evaluation score, the second evaluation score, and the composite evaluation score correspond to the evaluation score according to the present invention.

The present invention is not limited to the embodiments described above, and can be modified as appropriate without departing from the spirit of the present invention. For example, in the embodiment, two high performers are selected, but the number of high performers can be set as appropriate.
In addition, in the example, high performers were selected from among employees who participated in the same training, but if the training content is the same, human resources evaluation would be based on high performers selected from among employees who participated at other training venues. You can go. For example, if training with the same content is held at the Tokyo venue and the Osaka venue, the high performers selected at one venue can be used as the standard for talent evaluation of employees who participated at the other venue. It is preferable that the trainings conducted at these different venues be conducted at the same time as much as possible.
Further, in the embodiment, Sentence-Bert is used for natural language processing and vectorization processing, but the present invention is not limited to this, and other known processing may also be used.
Further, in the embodiment, training for corporate employees is exemplified, but the present invention is not limited to this, and can be similarly implemented for training for members of unions or organizations, training (courses) for school students, etc.

1 Human resource evaluation system 15 Recovery storage unit (recovery storage means)
21 Evaluation score acquisition means 22 Vectorization processing means 23 Similarity calculation means

Claims (2)

A human resources evaluation system that evaluates multiple participants who participated in a prescribed training,
an answer storage means for storing answer texts inputted by the participants in response to specific questions set in advance regarding the training in association with each participant;
An evaluation score for evaluating each participant is calculated based on the proportion of each participant's answer text related to the answer text of a specific participant selected in advance from among the participants who participated in the training. and evaluation score acquisition means,
The evaluation score acquisition means includes:
vectorization processing means for analyzing the answer sentences of each participant who participated in the training by natural language processing and converting them into sentence vectors based on the analysis to obtain a sentence vector for each answer sentence;
The degree of cosine similarity between the sentence vector of each participant obtained by the vectorization processing means and the sentence vector of the specific participant is calculated, and the evaluation score is obtained based on the cosine similarity of each participant. A human resources evaluation system characterized by comprising: a degree calculation means. At least two specific questions are set as the specific questions,
The vectorization processing means obtains a respective sentence vector from the answer sentence of each specific question for each participant,
The similarity calculation means includes:
processing means for synthesizing the sentence vectors obtained from the answer sentences of each specific question for each participant to obtain a composite sentence vector ;
and processing means for calculating the cosine similarity between the composite sentence vector of each participant and the composite sentence vector of the specific participant, and obtaining the evaluation score of each participant based on the cosine similarity. The human resources evaluation system according to claim 1, characterized in that:

Images