KR102677843B1 - Device of Customized Career Prediction in Life Cycle of Atheletes, and Career Prediction method using the same, and a computer-readable storage medium - Google Patents

Abstract

The present invention provides a career prediction device customized for each athlete's life cycle, comprising: a data storage unit storing athlete information; An embedding unit that extracts the athlete's award performance information and history information from the athlete information and generates a word embedding vector, and inputs the word embedding vector into a CNN (Convolutional Neural Network) model, which is a convolutional neural network, to generate the athlete's career information. It is possible to provide a career prediction device customized for each athlete's life cycle, including a career prediction unit, wherein the career information is a probability value for at least one career path that the athlete can advance into.

Description

A computer-readable medium in which a customized career prediction device for each athlete's life cycle, a career prediction method using the same, and a computer program for providing the same are recorded {Device of Customized Career Prediction in Life Cycle of Atheletes, and Career Prediction method using the same, and a computer-readable storage medium}

The present invention relates to a career prediction device customized for each athlete's life cycle, a career prediction method using the same, and a computer-readable medium on which a computer program for providing the same is recorded. More specifically, to predict the career path of an athlete, the present invention relates to a career prediction device tailored to each athlete's life cycle. A customized career prediction device for each athlete's life cycle that analyzes the awards, competition participation performance and history and predicts the appropriate career path for each individual athlete's life cycle based on the past athletes' career paths, a career prediction method using the same, and a method for providing the same. It relates to a computer-readable medium on which a computer program is recorded.

Although the number of professional athletes is rapidly increasing, due to the characteristics of professional athletes, with the exception of a few successful athletes, many athletes experience significant difficulties in career path and career development. Accordingly, an appropriate individual career guide is needed that takes into account the athlete's sport, personal characteristics, and history. To achieve this, it is necessary to provide career guidance to individual athletes through career information of past athletes based on their awards and track records.

However, the existing sports-related career guidance system only recommends appropriate sports based on surveys, and athletes' awards and records are recorded in natural language, making quantitative analysis difficult, and there are more than dozens of sports sports. The history information between these stocks has different problems.

Additionally, recommending an appropriate career path by referring to the vast information of thousands of past athletes requires considerable time and effort. Except for a small number of athletes who choose to pursue a career in the field of physical education and become successful as national team members or professional athletes, many athletes are experiencing difficulties in their career path, so an appropriate career guide that takes into account the careers of athletes is needed, and an efficient career guide is needed. For this purpose, a technology is needed that can automatically analyze the history of different athletes for each sport and, based on this, predict career paths through information on past athletes.

Meanwhile, deep learning technology, one of the recent artificial intelligence technologies, showed excellent performance in several fields (e.g. finance, weather, etc.).

Accordingly, there is a need for research that can predict and guide the career paths of athletes using deep learning technology and data from professional and non-professional athletes for about 100 sports accumulated over a long period of time.

Republic of Korea Patent No. 10-2143586 (2020.08.05)

In order to solve the above problems, the present invention relates to a career prediction device customized for each athlete's life cycle, a career prediction method using the same, and a computer-readable medium on which a computer program for providing the same is recorded, and the individual history and awards of athletes In order to analyze performance and recommend an appropriate career path based on the career paths of past athletes, the purpose is to provide guidance by analyzing the history and awards of individual athletes and recommending appropriate career paths based on the career paths of past athletes.

In order to solve the above problems, a customized career prediction device for each athlete's life cycle according to an embodiment of the present invention includes a data storage unit storing athlete information; An embedding unit that extracts the athlete's award performance information and history information from the athlete information and generates a word embedding vector, and inputs the word embedding vector into a CNN (Convolutional Neural Network) model, which is a convolutional neural network, to generate the athlete's career information. It includes a career prediction unit, and the career information is characterized in that it is a probability value for at least one career path that an athlete can advance into.

In addition, the embedding unit includes an embedding learning unit that generates and learns a corpus embedding vector based on a sports domain corpus; A refined information generator that generates refined information by refining the extracted award performance information and history information to include only substantive morphemes, and an embedding extractor that generates the word embedding vector from the refined information based on the learned corpus embedding vector. It can be included.

In addition, the embedding unit extracts syllable-level words from the sports domain corpus and the refined information using fastText and then generates the corpus embedding vector and the word embedding vector.

In addition, the career prediction unit may include a preprocessor that adds a token that divides the athlete's history information into months before inputting the word embedding vector into the CNN model and applies the word embedding vector to the token.

In addition, the preprocessor is characterized by dividing the history information of athletes whose life cycles have been completed into year cycles in order to generate a plurality of training data.

In addition, the career prediction method using a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention includes a data storage step of storing athlete information in a data storage unit; An embedding step of generating a word embedding vector by extracting the athlete's award performance information and history information from the data storage unit, and inputting the word embedding vector into a CNN (Convolutional Neural Network) model, which is a convolutional neural network, to generate the athlete's career information. It is possible to provide a customized career prediction method for each athlete's life cycle, including the career prediction stage.

In addition, the embedding step includes a learning step of generating and learning a corpus embedding vector based on the sports domain corpus; A refined information generation step of generating refined information by refining the extracted award performance information and history information to include only substantive morphemes, and an embedding extraction step of generating the word embedding vector from the refined information based on the learned corpus embedding vector. May include steps.

In addition, in the career prediction step, before inputting the word embedding vector into the CNN model, a token that divides the athlete's history information by month is added, the word embedding vector is applied to the token, and a plurality of training data are used. In order to generate it, a pre-processing step may be included in which the history information of the athlete whose life cycle is completed is divided into year cycles.

In addition, according to an embodiment of the present invention, it is possible to provide a computer-readable recording medium on which a computer program that performs a customized career prediction method for each athlete's life cycle is recorded.

The career prediction device customized for each athlete's life cycle according to the embodiment of the present invention as described above, the career prediction method using the same, and the computer-readable medium on which the computer program for providing the same are recorded take into account the athlete's sports and the individual's history. By understanding information and providing customized education and career guidance, we help athletes develop optimal careers.

In addition, the effects according to the embodiments of the present invention mentioned above are not limited to the contents described, and may further include all effects predictable from the specification and drawings.

Figure 1 is a flowchart of a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention.
Figure 2 is a flowchart of the embedding extraction unit of the customized career prediction device for each athlete's life cycle according to an embodiment of the present invention.
Figure 3 is an exemplary diagram illustrating the progress stages of a customized career prediction device for each athlete's life cycle according to an embodiment of the present invention.
Figures 4 and 5 are tables showing the evaluation values of the predicted career with respect to the actual career when the career of an athlete is predicted through a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention.
Figure 6 is a flow chart sequentially showing a customized career prediction method for each athlete's life cycle according to an embodiment of the present invention.
FIG. 7 is a flowchart schematically showing steps that can be performed in step S20 of FIG. 6.

Hereinafter, the description of the present invention with reference to the drawings is not limited to specific embodiments, and various changes may be made and various embodiments may be possible. In addition, the content described below should be understood to include all conversions, equivalents, and substitutes included in the spirit and technical scope of the present invention.

In the following description, the terms first, second, etc. are terms used to describe various components, and their meaning is not limited, and is used only for the purpose of distinguishing one component from other components.

Like reference numerals used throughout this specification refer to like elements.

As used herein, singular expressions include plural expressions, unless the context clearly dictates otherwise. In addition, terms such as “comprise,” “provide,” or “have” used below are intended to designate the presence of features, numbers, steps, operations, components, parts, or a combination thereof described in the specification. It should be construed and understood as not precluding the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

In addition, terms such as "unit", "unit", and "module" used in the specification refer to a unit that processes at least one function or operation, which is implemented as hardware, software, or a combination of hardware and software. It can be.

Hereinafter, with reference to the attached drawings, we will look in detail at a career prediction device customized for each life cycle of an athlete according to an embodiment of the present invention, a career prediction method using the same, and a computer-readable medium on which a computer program for providing the same is recorded. .

For reference, in the following description, athletes are used as an example for easy understanding of the present invention, but embodiments of the present invention can be applied in the same way to other occupational groups other than athletes.

Figure 1 is a flowchart of a career prediction device customized for each life cycle of an athlete according to an embodiment of the present invention, Figure 2 is a flowchart of the embedding extraction unit of a career prediction device customized for each life cycle of an athlete according to an embodiment of the present invention, and Figure 3 is This is an exemplary diagram showing the progress stages of a customized career prediction device for each athlete's life cycle according to an embodiment of the present invention.

Referring to Figures 1 to 3, the career prediction device 10 customized for each athlete's life cycle according to an embodiment of the present invention is a system that predicts the athlete's career using the athlete's award performance and history information, and includes a data storage unit ( 100), an embedding unit 200, and a path prediction unit 300.

The data storage unit 100 stores all data related to the athlete who is the subject of career prediction, and may include the athlete's individual resume, award information, articles related to the athlete, etc.

Referring to FIG. 2, the embedding unit 200 extracts the athlete's award performance information and history information from the data storage unit 100 and generates a word embedding vector, including the embedding learning unit 210 and the refined information generating unit. It may include 220 and an embedding extraction unit 230.

First, the embedding learning unit 210 can generate and learn a corpus embedding vector based on the sports domain corpus.

More specifically, the embedding learning unit 210 can collect sports domain corpus by extracting articles for each sport from newspaper articles in the 'National Institute of the Korean Language Everyone's Corpus'. Thereafter, the embedding learning unit 210 can generate processed data by processing the sports domain corpus for embedding learning using the fastText method, a sub-word expression model that generates word expressions on a syllable basis.

At this time, the processed data extracts a list of words to learn from the sports domain corpus and may be sub-word information for each extracted word.

Accordingly, the embedding learning unit 210 can generate a vectorized corpus embedding vector by learning words to be learned through word embedding learning using a learning model consisting of an input/output layer and a projection layer using processed data.

Here, the fastText method is a library for learning word embedding and text classification, and is one of the methods of turning words into vectors. More specifically, the fastText method considers that multiple words exist within one word, and can perform learning considering multiple words within a word.

The refined information generation unit 220 extracts the athlete's award performance information and history information from the data storage unit 100 and then refines them to generate refined information.

At this time, the award information may include all information about awards related to sports competitions, such as “winner in youth competition,” “winner of professional league,” and “winner of amateur team league,” while history information may include “joined professional team,” “amateur team league winner,” etc. It can include all information related to the athlete’s career, such as “appointment of team manager.”

It is most desirable to use a Korean morpheme analyzer to refine the data of the award performance information and history information extracted as above so that proper nouns (names of people, team names, schools, etc.) can be preserved.

Here, there are various types of Korean morpheme analyzers such as Kiwi, HAM, HLX, and Mecab. It is most preferable to use the Kiwi morpheme analyzer, but it is not limited to this.

More specifically, you can use the Kiwi morpheme analyzer to perform morpheme analysis on award performance information and history information and use the results.

Accordingly, the refined information generation unit 220 removes particles such as 'eul' and 'eul' or endings such as 'do' and 'was' included in the award performance information and history information, so that it does not have a word embedding vector in the future. You can avoid it. That is, the refined information generator 220 has the advantage of removing unnecessary information in advance by ensuring that the refined information includes only substantive morphemes such as common nouns, verbs, and adjectives.

The embedding extraction unit 230 may generate a word embedding vector from the refined information.

At this time, the embedding extraction unit 230 extracts low-order word refinement information for words included in the refinement information using the fastText method as in the embedding learning unit 210, and then extracts the extracted low-word refinement information. You can create a word embedding vector from .

Accordingly, the embedding unit 200 learns a corpus embedding vector through the embedding learning unit 210, and generates a word embedding vector using the athlete's award performance information and history information based on the learned corpus embedding vector to create a career prediction unit. (300) can be used to predict the athlete's career path.

The career prediction unit 300 may generate the athlete's career information by inputting the word embedding vector into a CNN (Convolutional Neural Network) model, which is a convolutional neural network.

Here, the athlete's career information refers to the probability value of at least one career path that the athlete can advance into. At this time, career paths related to athletes may be one or more of retirement, administrator, leader, national representative, youth representative, and referee, but are not limited to this and may include all career paths related to athletes.

More specifically, the career prediction unit 300 acquires a learning model for the correlation between the corpus embedding vector of the embedding learning unit 210 and the word embedding vector of the embedding extraction unit 230 through a CNN (Convolutional Neural Network) model. can do.

Accordingly, the career prediction unit 300 may generate career information by performing segmentation and combination analysis on the input word embedding vector for the purpose of generating career information for athletes based on the learning model.

At this time, the career path prediction unit 300 may be performed through the Input Layer, Convolutional Layer, Dropout Layer, Hidden Layer, and Output Layer of a CNN (Convolutional Neural Network) model.

The Input Layer can receive word embedding vectors and transmit them to the Convolutional Layer.

Convolutional Layer can analyze data by dividing and combining word embedding vectors into a specific number of consecutive words. In this case, the results derived from the convolutional layer may be the result of analyzing partial word sets composed of different numbers of words.

Dropout Layer may drop some information from analysis results. Through this elimination process, the quality of the analysis results can be improved by preventing the overfitting phenomenon in which the artificial intelligence analysis model is excessively adjusted to the learning data and by adding the randomness of unstructured text that can occur in reality.

Afterwards, the data is sequentially processed by the Hidden Layer and the Output Layer, and a single numerical information for generating career information is derived through the Output Layer. In this case, an iterative learning process may be performed in which the probability value given to the career information is compared and the degree of error is reflected in the analysis model. While the entire learning process is being performed, iterative learning can be performed on word embedding vectors.

At this time, the career prediction unit 300 may learn by setting the initial probability value of each piece of career information to 1.

Here, the initial probability value is a value expressed in the form of a one-hot vector, where one element can be expressed as 1 and the rest as 0, and each value can have the meaning of 100% and 0%. .

The career prediction unit 300 of the present invention learns by setting the probability value for the correct answer to 1 in the form of a one-hot vector for each piece of career information, thereby taking into account the fact that athletes can advance into various career paths simultaneously.

For example, when career information consists of [national representative, leader, administrator], and if the athlete in question became a leader after serving as a national representative, the initial probability value can be assigned as [1, 1, 0] and then learned.

Accordingly, the career prediction unit 300 is not limited to one type of career information for an athlete, but derives all career information suitable for the athlete, allowing the athlete to expand the range of choices.

In addition, the career prediction unit 300 adds a token that divides the athlete's history information into 6-month units before inputting the word embedding vector into the CNN model, and may include a preprocessor that applies the word embedding vector to the token. .

At this time, the preprocessor may divide the history information of the athlete whose life cycle has been completed into year cycles in order to generate a large number of training data.

In this way, the career prediction unit 300 divides the history information into year cycles and then classifies the history information into months through tokens, so that the athlete's history information can be clearly classified according to the passage of time. At this time, it is most desirable for the career prediction unit 300 to divide the history information into 3-year cycles and classify tokens into 6-month increments, but it is not limited to this. Through this, the career prediction unit 300 has the advantage of generating career information more quickly and accurately when predicting the career of an athlete.

The career prediction device 10 customized for each athlete's life cycle according to an embodiment of the present invention can generate the athlete's career information according to the steps described above.

For example, referring to Figure 3, through the above steps, award information and history information related to athletes are written in newspaper articles or various articles such as 'Participation in the 3rd Talent National Middle School Volleyball Competition in Seoul in August 2011'. is extracted, the extracted information is embedded and expressed as a vector, and then the athlete's career path can be predicted using the vector, such as 'probability of national team 42%, probability of leader 13%, etc.'

Figures 4 and 5 are tables showing the evaluation values of the predicted career with respect to the actual career when the athlete's career is predicted through a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention.

Referring to Figures 4 and 5, when an athlete's career is predicted through a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention, it is possible to check how well the athlete has matched the actual career path.

At this time, the career paths of athletes were divided into six categories (retired, administrator, leader, national representative, youth representative, referee), and there were three performance evaluation scales: 1-best accuracy, Recall@2, and Full accuracy.

1-best accuracy selects the one that the model predicts most strongly out of a total of six categories and evaluates whether it was on the actual course. In Recall@2, the model predicts two paths out of six categories and evaluates whether any of the two paths were actually on the path. Full Accuracy predicts one or more career paths out of six categories and evaluates whether the predicted career path is completely accurate with the actual career path. 1-best accuracy can be calculated through Equation 1, Recall@2 can be calculated through Equation 2, and Full accuracy can be calculated through Equation 3.

[Formula 1]

At this time, the label is the actual career path of the athlete. 1-best accuracy is Correct predictions divided by All predictions. All predictions of the model are All predictions, and Correct predictions are when one career path predicted by the model is included in the actual career path.

[Formula 2]

At this time, the label is the actual career path of the athlete. Recall@2 accuracy is the value divided by Correct recalls by All predictions. All predictions of the model are All predictions, and Correct recall@2 is when the model predicts two things and at least one of them is on the actual path.

[Formula 3]

At this time, the label is the actual career path of the athlete. Full-accuracy is the value of Correct full predictions divided by All predictions. All predictions of the model are All predictions, and Correct full prediction is when the N numbers predicted by the model are exactly the same as the label.

The performance of the career prediction device 10 customized for each athlete's life cycle according to an embodiment of the present invention was measured using the above evaluation indicators. To verify the performance of the sports domain embedding model technique, we verified the cases where sports domain embedding was not used, when the word2vec method, a word representation model, was used, and when the fastText method, a subword representation model, was used. Additionally, for each embedding model, LSTM, a recurrent neural network model, and CNN, a convolutional neural network model, were used for career prediction.

Figure 4 shows the results of measuring the career prediction performance of volleyball athletes through a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention. This is the result of measuring the performance of a combination of a sports domain embedding model and a career prediction model. From the table, you can see that career prediction can be done more effectively when embedding with the fastText method, a subword expression model, and using CNN, a convolutional neural network.

Figure 5 shows the results of measuring the career prediction performance of speed skating athletes through a career prediction device customized for each athlete's life cycle according to an embodiment of the present invention. This is the result of measuring the performance of a combination of a sports domain embedding model and a career prediction model. From the table, you can see that career prediction can be done more effectively when embedding with the fastText method, a subword expression model, and using CNN, a convolutional neural network.

This customized career prediction method for each athlete's life cycle will be explained in detail below.

Figure 6 is a flowchart sequentially showing a customized career prediction method for each athlete's life cycle according to an embodiment of the present invention, and Figure 7 is a flowchart schematically showing steps that can be performed in step S20 of Figure 6.

Referring to FIG. 6, the customized career prediction method for each athlete's life cycle according to an embodiment of the present invention may include a data storage step (S10), an embedding step (S20), and a career prediction step (S30).

In the data storage step (S10), athlete information can be stored in the data storage unit 100 to form a dataset.

In the embedding step (S20), a word embedding vector may be generated by extracting the athlete's award performance information and history information from the data storage unit 100.

Referring to FIG. 7, the embedding step (S20) may include a learning step (S21), a refinement information generation step (S22), and an embedding extraction step (S23). Since each step constituting the embedding step (S20) has been described in the above device, detailed description will be omitted.

The learning step (S21) can be learned by generating a corpus embedding vector based on the sports domain corpus.

In the refined information generation step (S22), the athlete's award performance information and history information are extracted from the data storage unit 100, and then the award performance information and history information are refined to include only substantive morphemes using a Kiwi morpheme analyzer. can be created.

The embedding extraction step (S23) can generate a word embedding vector by matching the purification information with the corpus embedding vector.

Lastly, in the career prediction step (S30), the athlete's career information can be generated by inputting the word embedding vector into a CNN (Convolutional Neural Network) model.

More specifically, the career prediction step (S30) performs segmentation and combination analysis of input word embeddings for the purpose of generating career information for athletes through a CNN (Convolutional Neural Network) model and identifies patterns in the input data. Thus, the probability value for the career path can be derived.

At this time, the career path prediction step (S30) may include a preprocessing step performed before inputting the word embedding vector into the CNN model.

In the preprocessing step, before finding a pattern, tokens that divide the athlete's history information by months are added, the word embedding vector is applied to the token, and the history information of the athlete whose life cycle is completed is generated to generate a large number of training data. can be divided into year cycles.

The customized career prediction device for each athlete's life cycle according to the embodiment of the present invention described above can also be implemented in the form of a recording medium containing instructions executable by a computer, such as a program module executed by a computer. Such recording media includes computer-readable media, which can be any available media that can be accessed by a computer and includes both volatile and non-volatile media, removable and non-removable media. Computer-readable media also includes computer storage media, both volatile and non-volatile implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. , includes both removable and non-removable media.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concept of the present invention defined in the following claims are also possible. It falls within the scope of rights.

10: Customized career prediction device for each athlete’s life cycle
100: data storage unit
200: Embedding part
210: Embedding learning unit
220: Purification information generation unit
230: Embedding extraction unit
300: Career prediction department

Claims (9)

A data storage unit storing athlete information;
An embedding unit that generates a word embedding vector by extracting the athlete's award performance information and history information from the athlete information, and
A career prediction unit that generates career information of the athlete by inputting the word embedding vector into a CNN (Convolutional Neural Network) model, a convolutional neural network,
The above career information is,
It is the probability value for at least one career path that an athlete can advance into,
The career prediction unit,
Before inputting the word embedding vector into the CNN model, a preprocessor adds a token that divides the athlete's history information into months and applies the word embedding vector to the token,
The preprocessor,
In order to generate a large number of training data, the above history information of athletes whose life cycle has been completed is divided into year cycles,
The career prediction unit divides the history information into year cycles and then classifies the history information into months through the token,
The embedding part,
An embedding learning unit that generates and learns corpus embedding vectors based on the sports domain corpus;
A refined information generation unit that refines the extracted award performance information and history information to include only substantive morphemes to generate refined information;
An embedding extraction unit that generates the word embedding vector from the refined information based on the learned corpus embedding vector,
The embedding part,
Characterized by extracting syllable-level words from the sports domain corpus and the refined information using fastText, and then generating the corpus embedding vector and the word embedding vector,
The embedding learning unit,
Processed data is generated by processing the sports domain corpus for embedding learning using the fastText method, which is a sub-word expression model that generates word expressions on a syllable basis, and the processed data is words extracted from the sports domain corpus. It consists of subword information about the list,
The embedding learning unit learns the processed data through word embedding learning to generate the corpus embedding vector,
The embedding extractor,
Extracting low-level word refinement information for words included in the refining information using the fastText method and generating the word embedding vector from the extracted low-word refining information,
The career prediction unit,
A customized career prediction device for each athlete's life cycle, where the initial probability value for each piece of career information is set and learned in the form of a one-hot vector, where at least one element is expressed as 1 and the rest are expressed as 0.
delete delete delete delete In the customized career prediction method for athletes in a customized career prediction device for each athlete's life cycle,
A data storage step of storing athlete information in a data storage unit;
An embedding step in which the embedding unit extracts the athlete's award performance information and history information from the athlete information and generates a word embedding vector;
It includes a career prediction step in which the career prediction unit inputs the word embedding vector into a CNN (Convolutional Neural Network) model, which is a convolutional neural network, to generate career information for the athlete,
The career prediction step is,
Before inputting the word embedding vector into the CNN model, tokens that separate the athlete's history information by months are added, the word embedding vector is applied to the token, and the life cycle is completed to generate a plurality of training data. Including a pre-processing step of dividing the athlete's history information into year cycles,
The career prediction unit divides the history information into year cycles and then classifies the history information into months through the token,
The embedding step is,
A learning step of generating and learning a corpus embedding vector based on the sports domain corpus;
A refined information generation step of generating refined information by refining the extracted award performance information and history information to include only substantive morphemes;
An embedding extraction step of generating the word embedding vector from the refined information based on the learned corpus embedding vector,
The embedding step is,
Characterized by extracting syllable-level words from the sports domain corpus and the refined information using fastText, and then generating the corpus embedding vector and the word embedding vector,
The learning stage is,
Processed data is generated by processing the sports domain corpus for embedding learning using the fastText method, which is a sub-word expression model that generates word expressions on a syllable basis, and the processed data is words extracted from the sports domain corpus. It consists of subword information about the list,
The learning step is to generate the corpus embedding vector by learning the processed data through word embedding learning,
The embedding extraction step is,
Extracting low-level word refinement information for words included in the refining information using the fastText method and generating the word embedding vector from the extracted low-word refining information,
The career prediction step is,
A customized career prediction method for each athlete's life cycle in which the initial probability value for each piece of career information is set and learned in the form of a one-hot vector, where at least one element is expressed as 1 and the rest are expressed as 0.
delete delete A computer-readable recording medium on which a computer program that performs a customized career prediction method for each athlete's life cycle according to paragraph 6 is recorded.