KR20240021387A - Job search matching method and system - Google Patents

Abstract

The present invention relates to a matching method and system that can provide high-quality company recommendation services to users based on job postings on a job search site. It matches and provides job postings that meet the user's requirements based on the user's selection. It relates to a method and system that can do this, comprising: receiving at least one input of corporate conditions and user specifications desired by the user; learning and calculating a correlation between companies satisfying the company conditions and company preferences according to at least one user specification; determining the ranking of companies satisfying the company conditions according to the correlation; and presenting a list of the companies to the user according to the ranking of the companies according to the correlation.

Description

Job matching method and system{JOB SEARCH MATCHING METHOD AND SYSTEM}

The present invention relates to a matching method and system that can provide high-quality company recommendation services to users based on job postings on a job search site. It matches and provides job postings that meet the user's requirements based on the user's selection. It's about methods and systems that can do it.

In the past, you could get information through the job search section of newspapers or through publications, but today, you can check job openings at various companies through the Internet or app services and check your application qualifications. You can select and apply to companies that meet your desired conditions.

Company recommendations from existing job search sites are made in the order of company list by category, reviews, benefits, and salary. Therefore, in order to select a company that the user wishes to apply to, the user must check and select information such as company list by category, reviews, benefits and salary, etc. In order to select a company to apply among numerous companies, the user must select the company's information. This consumes a considerable amount of time because you have to check them one by one to make sure they meet the conditions you want.

Therefore, if similar companies are recommended through machine learning for a job posting that the user wants, it will be easier for the user to use it than directly entering various conditions into the system. There is a need for a system that provides satisfactory recommendation results to users while being as simple and intuitive as possible.

Therefore, the present invention is intended to solve the problems described above, and seeks to provide a method and system for listing and recommending companies that suit the user's preferences.

The present invention seeks to provide a method and system that can increase accuracy while reducing the burden on the system by finding attributes that affect the user's achievement of the goal and minimizing the amount of calculation.

The present invention seeks to provide a method and system that uses various visualization methods to enable users to understand the characteristics of corporate job postings using only visualization data without necessarily reading the text.

A job matching method according to an embodiment of the present invention to solve the above problem includes receiving input of at least one company condition and user specification desired by the user; learning and calculating a correlation between companies satisfying the company conditions and company preferences according to at least one user specification; determining the ranking of companies satisfying the company conditions according to the correlation; and presenting a list of the companies to the user according to the ranking of the companies according to the correlation.

According to one embodiment of the present invention, the correlation calculation step includes matching at least one user specification and company preference; learning the correlation between the change in the user specifications and the change in the company's preference by defining it as a probability; and calculating a correlation between the company conditions and company preferences according to at least one user specification according to the probability.

According to an embodiment of the present invention, in the correlation calculation step, the correlation may be calculated using Equation 1 below.

[Equation 1]

(Here, C(S, P) is the correlation, E(S) is the analysis relevance value, S is the user specification, and P is the company preference.)

According to an embodiment of the present invention, the user specifications may include at least one of school, major, gender, age, region, and MBTI.

According to one embodiment of the present invention, the step of determining the ranking of companies determines a ranking including company preference according to a combination of a plurality of user specifications, and the user presentation step provides a result according to the ranking. You can.

A job matching system according to an embodiment of the present invention to solve the above problem includes a data processing unit that receives at least one input of company conditions and user specifications desired by the user; a learning unit that learns and calculates a correlation between companies that satisfy the company conditions and company preferences according to at least one user specification; a ranking unit that determines the ranking of companies that satisfy the company conditions according to the correlation; and a visualization unit that presents a list of companies to the user according to the ranking of the companies according to the correlation.

According to one embodiment of the present invention, the learning unit matches at least one user specification and corporate preference, defines the correlation between the change in the user specification and the change in the corporate preference as a probability, and learns according to the probability. The correlation between the corporate conditions and corporate preference according to at least one user specification can be calculated.

According to an embodiment of the present invention, the learning unit may calculate the correlation using Equation 1 below.

[Equation 1]

(Here, C(S, P) is the correlation, E(S) is the analysis relevance value, S is the user specification, and P is the company preference.)

According to an embodiment of the present invention, the user specifications may include at least one of school, major, gender, age, region, and MBTI.

According to an embodiment of the present invention, the learning unit determines a ranking including company preference according to a combination of a plurality of user specifications, and the visualization unit may provide a result according to the ranking.

According to the present invention, companies that fit the user's preferences can be listed and recommended by ranking.

In addition, it is possible to minimize the amount of computation by finding attributes that affect the user's goal achievement, thereby reducing the burden on the system and increasing the accuracy of predictions.

In addition, by using various visualization methods, users can understand the characteristics of company job postings using only visualization data without necessarily reading the text.

Meanwhile, the effects of the present invention are not limited to the effects mentioned above, and various effects may be included within the range apparent to those skilled in the art from the contents described below.

Figure 1 is a block diagram of a job matching system according to an embodiment of the present invention.
Figure 2 is a screen during analysis of a matching page applying the job search matching method according to an embodiment of the present invention.
Figure 3 is a matching result screen applying the job search matching method according to an embodiment of the present invention.
Figure 4 is a visualization result screen applying the job search matching method according to an embodiment of the present invention.
Figure 5 is a result screen when there is no desired information on a matching page applying the job search matching method according to an embodiment of the present invention.
Figure 6 is a result screen when there are no matching companies as a result of applying the job search matching method according to an embodiment of the present invention.
Figure 7 is a matching result screen applying the job search matching method according to an embodiment of the present invention.
Figure 8 is a flowchart of a job search matching method according to an embodiment of the present invention.

Hereinafter, the 'job search matching method and system' according to the present invention will be described in detail with reference to the attached drawings. The described embodiments are provided so that those skilled in the art can easily understand the technical idea of the present invention, and the present invention is not limited thereto. In addition, the matters expressed in the attached drawings may be different from the actual implementation form in the schematic drawings to easily explain the embodiments of the present invention.

Meanwhile, each component expressed below is only an example for implementing the present invention. Accordingly, other components may be used in other implementations of the present invention without departing from the spirit and scope of the present invention.

Additionally, each component may be implemented purely through hardware or software configurations, but may also be implemented through a combination of various hardware and software configurations that perform the same function. Additionally, two or more components may be implemented together by one hardware or software.

Additionally, the expression 'including' certain components is an 'open' expression and simply refers to the existence of the corresponding components and should not be understood as excluding additional components.

Figure 1 is a block diagram of a job matching system according to an embodiment of the present invention.

Referring to FIG. 1, the job matching system 1 according to an embodiment of the present invention may include a data processing unit 100, a learning unit 200, and a visualization unit 300.

The data processing unit 100 can receive input of at least one company condition and user specification desired by the user. The data processing unit 100 may collect information on job vacancies from companies through job vacancies, etc. The data processing unit 100 may crawl and collect job information for companies from job search sites. The data processing unit 100 may receive input of at least one company condition desired by the user. The data processing unit 100 may receive input of at least one specification of the user.

The data processing unit 100 crawls using a job search site (e.g., Job Planet) platform that has standardized job advertisement data among various job search sites in order to prepare standardized data on job search information. can do. Although the information provided by Job Planet is relatively standardized compared to other job search platforms, the data processing unit 100 can perform crawling appropriate to the situation by processing exceptions for job advertisements with insufficient information. The data processing unit 100 can perform data crawling through Selenium, a browser automation tool, and can collect job advertisements for all jobs over 2 to 3 days to collect a large amount of data.

The data processing unit 100 has a unified format, but can correct (stopword processing) different speech styles and unnecessary information for each job posting, and has a data tokenization step to extract meaningful data and increase learning accuracy. You can. According to one embodiment of the present invention, the data processing unit 100 uses konlpy (Mecab, Kkma, Okt, etc.), a popular Korean morpheme analyzer, and Kiwi tokenizer, the latest technology, to find a tokenization method suitable for the project. After directly comparing and analyzing, you can adopt Kiwi Tokenizer, which uses a more diverse part-of-speech tagging method. In conclusion, an organic tokenization method was used by mixing Kiwi+mecab tokenizer, and Mecab tokenizer was used for Korean and English combinations that could not be recognized by kiwi tokenizer. The data processing unit 100 may store the tokenized data as data for welfare benefits and qualification requirements. Later, the learning unit 200 can set up a learning model so that a user can be recommended a similar company by selecting one of two criteria for a similar job.

If the data processing unit 100 considers a document with the company name and job as a tag and the job information as a corpus in the company's recruitment announcement, it vectorizes the document to learn the similarity of documents with the company name and job title. You can proceed. For this purpose, the data processing unit 100 may use Doc2Vec, a document embedding technology extended from the existing word embedding technology Word2Vec. To apply Doc2Vec to data, you can use gensim, a Python library that provides most of the convenience functions needed to convert natural language into vectors.

The data processing unit 100 refines the accuracy of data preprocessing and converts the system into a Flask framework so that the user can conveniently use the model learned through Doc2Vec (a vector model indicating similarity for each company's job posting). can be developed As a result of the visualization unit 300, the user can check well-organized job postings, visual data, and job posting similarity that can confirm the reliability of the data through the web.

According to one embodiment of the present invention, the data processing unit 100 crawls all job postings registered in Job Planet and uses Selenium's web to crawl dynamically implemented parts such as JavaScript of the Job Planet site. Drivers are available. Although Job Planet provides relatively more standardized information than other job search platforms, since the number of types of information provided is different for each company, data can be collected using exception processing.

When an exception occurs, the data processing unit 100 can crawl the page again. After data crawling, the data can be divided by company name, job posting title, deadline, recruitment job, experience, employment type, salary, skill, company introduction, main tasks, qualifications, preferential treatment, recruitment procedure, and welfare benefits and saved as a CSV file. You can.

The data processing unit 100 can collectively perform preprocessing (stopword processing + merging data with similar characteristics into one column) on the stored CSV files. For example, required skills, qualifications, and preferential treatment can be lumped into the ‘Qualifications’ column because they have the same nature. For example, the job posting title, job title, and main tasks have the same nature, so they can be grouped into the 'task' column.

Since the data processing unit 100 recognizes and stores the same job posting as a different job posting if the deadline is different, if the 'company name + job posting title' is the same, duplicate data can be excluded using the drop_duplicates of the pandas module. You can.

When processing stop words, the data processing unit 100 will also provide raw data of job postings to the user, so the original job postings can be stored in a separate column.

When saving CSV, the data processing unit 100 may replace \r with \n in a situation where a line is automatically skipped if the job posting contains the character "\r".

The data processing unit 100 may compare tokenization techniques using various tokenizers and then use the tokenizer that is best at tokenizing (extracting important words from sentences).

As a result of comparing Kiwi, Mecab, Kkma, Okt, etc., Kiwi and Mecab tokenizers divide part-of-speech tags into various parts and are most suitable for learning.

The data processing unit 100 can tokenize stopword-processed data into 'common nouns, proper nouns, English, and roots' using the kiwi tokenizer.

The kiwi tokenizer generates an exception when it encounters a combination of Korean and English whose meaning is unknown. For example, an exception may occur during rigging TA (which often appears in job postings at game developers).

The data processing unit 100 can tokenize data that causes an exception in kiwi using the Mecab tokenizer.

The data processing unit 100 can create two CSV files using the developed tokenizer. For example, you can create a CSV file that tokenizes benefits and task columns, and a CSV file that tokenizes qualifications and task columns.

The reason why the data processing unit 100 additionally tokenizes the task column when creating each csv file is because two learning models will be created. For example, you can create a model that learns welfare/qualification requirements + a model that learns tasks.

The reason why the data processing unit 100 uses two learning models is to recommend companies that have similar criteria in one of the 'welfare/qualification requirements' to the company entered by the user, and the 'welfare/qualification requirements' are If the job is similar but the hiring position is different, the recommendation has no meaning to the user.

Employers attempt to find jobs in their field of expertise, and generally the types of jobs for any specialty are limited.

The data processing unit 100 performs learning on the tokenized data using the Doc2Vec model. At this time, the hyperparameter of the Doc2Vec model can be adjusted to select the value with the best accuracy.

The reason why Doc2Vec was used as a machine learning method at this time was because it was judged that Doc2Vec would be more advantageous for learning in that it is faster than the document embedding method using TensorFlow and can be learned by adjusting several hyperparameters.

The data processing unit 100 can preferentially extract the index of companies that are more than 70% similar to the company entered by the user from the model that learned the task among the two learned models.

The reason why the data processing unit 100 set the lower limit of job similarity to 70% was because the amount of data used when training the job model (the model that learned the task column) was large (the 'job posting title' in the task column , job, and main task data), even if there was a 70% similarity, it was similar when comparing the actual data.

The data processing unit 100 extracts similar companies using the 'welfare/qualification requirements' model for companies extracted from the job model and outputs them in descending order of similarity. For example, ultimately, companies with more than 70% similar jobs and similar 'welfare/qualification requirements' can be extracted.

The learning unit 200 may learn and calculate the correlation between companies that satisfy the company conditions and company preferences according to at least one user specification. The learning unit 200 can select and list companies that satisfy the company conditions. The learning unit 200 may rank companies that the user may prefer according to the correlation of company preferences according to the user's specifications from the list of companies that satisfy the company conditions.

The learning unit 200 may match at least one user specification and company preference. The learning unit 200 may learn the correlation between the change in user specifications and the change in company preference by defining it as a probability. The learning unit 200 may calculate a correlation between the company conditions and company preferences according to at least one user specification according to the probability.

The learning unit 200 may measure the correlation by defining the correlation between the change in the weather state attribute and the change in the state of the object as a probability and learning.

The learning unit 200 may calculate the influence of companies satisfying the company conditions and company preferences according to at least one user specification.

The influence is a value that varies depending on how much influence a change in the user specification has on the company's preference and can be calculated using Equation 1 below.

[Equation 1]

Here, p(x) is the probability that a change in preference x will occur. As the probability of occurrence approaches 1, the influence approaches 0, requiring learning of a small number of cases, thus reducing the amount of computation. As the probability of occurrence approaches 0, the influence becomes infinitely greater, requiring learning of many cases, which may increase the amount of computation.

In the learning unit 200, the calculation load represents the expected value (average) of the influence. A large calculation load means a large average influence, and the greater the uncertainty, the more difficult it is to classify, so the one with the smallest calculation load is the top doctor. The amount of computation can be reduced by placing it at the decision node. The computational load can be obtained by calculating Equation 2 below.

[Equation 2]

Here, S refers to the collection of all events that have already occurred, and c may refer to the number of events.

In the learning unit 200, the correlation can be obtained by calculating Equation 3 below.

[Equation 3]

Here, C(S, P) is the correlation, E(S) is the analysis relevance value, S is the user specification, and P is the company preference. Here, P refers to a specific specification among user specifications, and it can be determined which specification has the lowest computational load (i.e., the largest amount of information acquisition) when classified.

The learning unit 200 may determine the ranking of the user specifications according to the correlation. The learning unit 200 may determine the ranking in descending order of correlation. The learning unit 200 may prioritize the user specifications with the low correlation in recommendation.

The user specifications may include at least one of school, major, gender, age, region, and MBTI.

The learning unit 200 may determine a ranking including company preference according to a combination of a plurality of user specifications.

The visualization unit 300 may present a list of companies to the user according to the ranking of the companies according to the correlation.

The visualization unit 300 may provide results according to the ranking of the correlation of corporate preferences according to the plurality of user specifications. The visualization unit 300 may provide results according to the ranking of correlations between corporate preferences according to a plurality of combinations of the user specifications.

The visualization unit 300 can visualize data and provide it to the user so that it is easy to visually check which parts are similar when the user is recommended a similar company and checks the information in the recommended company announcement. The visualization unit 300 can use the Wordcloud method to variously visualize key words that frequently appear in each company's announcement content. The visualization unit 300 displays similar companies on an additional web page using TSNE (a technique for reducing the dimension of high-dimensional vector data) on a two-dimensional coordinate plane, allowing users to explore similar companies in an intuitive and interesting way. Functions that can be used can also be provided.

The visualization unit 300 can output the system developed up to the 'data embedding' stage by mounting it on a web server using Flask. Data transmission between the Backend and FrontEnd can be performed by sending data to the front-end as a parameter when the back-end calls the render_template function that shows the HTML file to the client. Data received from the back-end can be displayed in components placed on HTML using JavaScript and jQuery at the front-end. Data transfer from the front-end to the web server can be implemented by using the get method of the form tag, or by including data in the URL called by the front-end and then processing it at the back-end.

The web UI can be constructed intuitively and simply using various elements of bootstrap (ex. progressbar, radio button, table, card).

If the user needs to wait for the system to perform an operation, such as when the system is in the process of learning something new or creating visualization data, a progressbar can be used to let the user know what operation is in progress.

The composition of HTML elements can be designed for placement and size using CSS grammar in the style tag of <head>.

In the case of HTML files showing visualization information, new information must be provided rather than cached data in the web browser, so use Cache-Control to prevent caching and add a random number to the URL when calling the visualization image. By adding it and sending it to the server, the changed image can be provided to the user even if the image file of the same name is overwritten with other visualization information.

The structure of an example web page is as follows.

Start page: hello.html,

Job posting link input page: apply.html,

Loading page: index.html,

Analysis result page: result.html,

Company job posting details page: linktoimg.html,

Page notifying that there is no 'welfare' information: nobokri.html,

A page indicating that there are no similar companies: nolist.html;

Page showing visualization information: display_plot.html

The user enters the Job Planet job posting link in apply.html and then enters what criteria (benefits or qualifications) they would like to receive recommendations for similar companies.

The visualization unit 300 extracts similar companies from the learned model according to the link and criteria entered by the user on the web server and provides them to the user in a table format through a table in result.html.

In the visualization unit 300, the user clicks [View Details] for each company in result.html and is provided with WordCloud results that analyze the keywords of the company's job posting and raw job posting content that has not been preprocessed.

If the visualization unit 300 commands a recommendation as welfare benefits without information on 'welfare benefits' in the link entered by the user, the system displays nobokri.html to instruct the user to re-enter the link. ask for

If there is no company similar to the company in the link entered by the user (when there is no company with similar job duties and standards), the visualization unit 300 notifies the user that there is no similar company by displaying nolist.html, and informs the user Although the 'input criteria' are not similar, other companies with similar 'jobs' are recommended.

In the visualization unit 300, the user can move from result.html to display_plot.html, which shows visualization information.

The visualization unit 300 uses TSNE coordinate data received from the server in display_plot and draws a scatter plot on HTML using Plotly.js.

The visualization unit 300 displays two plots at this time: a scatter plot visualizing companies with similar 'user-selected criteria' to the user input company and a scatter plot visualizing companies with similar 'jobs'.

By clicking on the marker of each plot, the user is moved to linktoimg.html to receive job posting information from the company corresponding to that marker.

The visualization unit 300 provides not only WordCloud results and raw job posting content in linktoimg.html, but also 'similarity' according to each standard so that users can directly check the reliability of the data recommended by the system.

Also, in linktoimg.html, depending on the marker the user clicks on the 'Visualization Results' page, if the red marker is clicked, it is notified that it is a job advertisement entered by the user, and if the blue marker is pressed, the job advertisement is notified that the job advertisement is a job advertisement entered by the user. They were asked to provide what criteria they analyzed as similar.

The embedding vector of each job posting is 200-dimensional high-dimensional data.

In order to provide visualized information to the user, the visualization unit 300 reduces data to two dimensions using a TSNE module and then provides visualization on a two-dimensional plane.

At this time, the visualization unit 300 does not visualize information on all companies, but only visualizes and provides companies that are a certain distance away from the user input company (most similar companies).

Among the visualized information, the visualization unit 300 is configured to prevent any two markers from appearing as one marker due to a very small difference in position between any two markers. When the distance between any two markers is very small, the visualization unit 300 selects one of the two markers. Perform visualization by slightly changing the position so that they do not overlap.

The visualization unit 300 creates and provides tokenized welfare/qualification data as a WordCloud module using keywords.

At this time, if the user selects 'welfare' as the standard for determining similarity, the WordCloud result created with 'welfare' tokenized data is shown, and if the user selects 'qualification requirements', the WordCloud result created with 'qualification requirement' tokenized data is displayed. shows.

Figure 2 is a screen during analysis of a matching page applying the job opening matching method according to an embodiment of the present invention, Figure 3 is a matching result screen applying the job opening matching method according to an embodiment of the present invention, and Figure 4 is This is a visualization result screen applying the job search matching method according to an embodiment of the present invention, and Figure 5 is a result screen when there is no desired information on the matching page applying the job search matching method according to an embodiment of the present invention, and Figure 6 is a result screen when there are no matching companies as a result of applying the job vacancy matching method according to an embodiment of the present invention, and Figure 7 is a matching result screen after applying the job vacancy matching method according to an embodiment of the present invention.

2 to 7, according to an embodiment of the present invention, the data processing unit 100 copies the URL of a job advertisement for a desired company from a job search site (for example, Job Planet) and enters it into the system. Then, when you select an item you like among the welfare benefits and qualifications of the relevant job posting, the job postings of the URL entered by the user are collected from the job search site, and the visualization unit 300 analyzes the company and the job openings. Similar companies can be printed on a web page in table format. While performing calculations through the model, a loading screen can be displayed to the user, and analysis results can be provided after the calculation is completed.

When clicking [View Details] in the 'Details' column of the companies shown in the table displayed in the visualization unit 300, the actual job posting content of the company and the WordCloud analysis results of the content analyzed using keywords are provided to the user. can do.

When you click [View visualization data] at the bottom of the table format analysis results, companies with similar [jobs] and [user-entered criteria] to the company entered by the user are created into two-dimensional visualization data. It can be provided to the user.

In the visualization data of the visualization unit 300, red dots may represent companies whose URLs are entered by the user, and blue dots may represent companies analyzed to be similar to the red dot companies.

Even when a dot is clicked in the above visualization data, the visualization unit 300 can provide the user with [View details] information about the company corresponding to the dot.

On the page that provides detailed information on company job postings, the visualization unit 300 can display the content of the actual job posting, keyword WordCloud analysis results, and percentage similarity to the company entered by the user.

If there is no 'welfare' information in the job advertisement of the URL entered by the user and the user requests an analysis result based on 'welfare', the visualization unit 300 notifies that there is no welfare information in the link, You can request that the user re-enter the URL.

If the URL entered by the user is new data that the model has not learned, it takes more time to perform additional learning, so the visualization unit 300 adjusts the loading time and indicates to the user what operation is being performed. You can.

If there are no similar companies as a result of the analysis (if there are no companies with similar job duties and user input criteria), the visualization unit 300 can notify the user of this. The visualization unit 300 may then provide a link at the bottom right to check companies with similar jobs. When the user clicks on the link, a list of similar companies is output in table format, but it is specified as [Companies with similar jobs] so that the user can recognize the criteria for the analysis.

Figure 8 is a flowchart of a job search matching method according to an embodiment of the present invention.

Referring to FIG. 8, the job matching method according to an embodiment of the present invention may include the step of receiving at least one company condition and user specification desired by the user (S110).

In step S110, the data processing unit 100 may receive at least one input of company conditions and user specifications desired by the user. The data processing unit 100 may collect information on job vacancies from companies through job vacancies, etc. The data processing unit 100 may crawl and collect job information from companies on job search sites. The data processing unit 100 may receive input of at least one company condition desired by the user. The data processing unit 100 may receive input of at least one specification of the user.

In step S110, the data processing unit 100 selects a job search site (e.g., Job Planet) that has data on standardized job advertisements among various job search sites to prepare standardized data on job search information. You can crawl using the platform. Although the information provided by Job Planet is relatively standardized compared to other job search platforms, the data processing unit 100 can perform crawling appropriate to the situation by processing exceptions for job advertisements with insufficient information. The data processing unit 100 can perform data crawling through Selenium, a browser automation tool, and can collect job advertisements for all jobs over 2 to 3 days to collect a large amount of data.

In step S110, the data processing unit 100 uses a data token to correct (process stopwords) the format and the different speech and unnecessary information for each job posting, even though the format is unified, and to extract meaningful data and increase the accuracy of learning. There may be stages of anger. According to one embodiment of the present invention, the data processing unit 100 uses konlpy (Mecab, Kkma, Okt, etc.), a popular Korean morpheme analyzer, and Kiwi tokenizer, the latest technology, to find a tokenization method suitable for the project. After directly comparing and analyzing, you can adopt Kiwi Tokenizer, which uses a more diverse part-of-speech tagging method. In conclusion, an organic tokenization method was used by mixing Kiwi+mecab tokenizer, and Mecab tokenizer was used for Korean and English combinations that could not be recognized by kiwi tokenizer. The data processing unit 100 may store the tokenized data as data for welfare benefits and qualification requirements. Later, the learning unit 200 can set up a learning model so that a user can be recommended a similar company by selecting one of two criteria for a similar job.

In step S110, if the data processing unit 100 considers the company's job announcement as a document with the company name and job as a tag and the job information as a corpus, the data processing unit 100 learns the similarity of documents containing the company name and job title. You can vectorize the document. For this purpose, the data processing unit 100 may use Doc2Vec, a document embedding technology extended from the existing word embedding technology Word2Vec. To apply Doc2Vec to our data, we can use gensim, a Python library that provides most of the convenience functions needed to convert natural language to vectors.

In step S110, the data processing unit 100 refines the accuracy of data preprocessing and configures the system so that the user can conveniently use the model learned through Doc2Vec (a vector model indicating similarity for each company's job posting). Developed with the Flask framework. As a result of the visualization unit 300, the user can check well-organized job postings, visual data, and job posting similarity that can confirm the reliability of the data through the web.

In step S110, according to one embodiment of the present invention, the data processing unit 100 crawls all job postings registered in Job Planet and crawls dynamically implemented parts such as JavaScript of the Job Planet site. To do this, you can use Selenium's web driver. Although Job Planet provides relatively more standardized information than other job search platforms, since the number of types of information provided is different for each company, data can be collected using exception processing.

In step S110, the data processing unit 100 may crawl the page again when an exception occurs. After crawling the data, the data can be divided by company name, job posting title, deadline, recruitment job, experience, employment type, salary, skill, company introduction, main tasks, qualifications, preferential treatment, recruitment procedure, and welfare benefits and saved as a CSV file. You can.

In step S110, the data processing unit 100 may collectively perform preprocessing (stopword processing + merging data with similar characteristics into one column) on the stored CSV files. For example, required skills, qualifications, and preferential treatment can be lumped into the ‘Qualifications’ column because they have the same nature. For example, the job posting title, job title, and main tasks have the same nature, so they can be grouped into the 'task' column.

In step S110, the data processing unit 100 recognizes and stores the same job posting as a different job posting if the deadline is different. Therefore, if the 'company name + job posting title' are the same, duplicates are detected using the drop_duplicates of the pandas module. Data can be excluded.

In step S110, when processing stop words, the data processing unit 100 will also provide raw data of the job posting to the user, so the original job posting can be created and stored in a separate column.

In step S110, the data processing unit 100 may replace \r with \n in a situation where a line is automatically skipped if the character "\r" is included in the job posting when saving CSV. .

In step S110, the data processing unit 100 may compare tokenization techniques using several tokenizers and then use the tokenizer that is best at tokenizing (extracting important words from sentences well).

As a result of comparing Kiwi, Mecab, Kkma, Okt, etc., Kiwi and Mecab tokenizers divide part-of-speech tags in the most diverse ways and are suitable for learning.

In step S110, the data processing unit 100 can tokenize stopword-processed data into 'common nouns, proper nouns, English, and roots' using the kiwi tokenizer.

The kiwi tokenizer generates an exception when it encounters a combination of Korean and English whose meaning is unknown. For example, an exception may occur during rigging TA (which often appears in job postings at game developers).

In step S110, the data processing unit 100 can tokenize data that causes an exception in kiwi using the Mecab tokenizer.

In step S110, the data processing unit 100 can create two csv files using the developed tokenizer. For example, you can create a CSV file that tokenizes benefits and task columns, and a CSV file that tokenizes qualifications and task columns.

In step S110, the data processing unit 100 additionally tokenizes the task column when creating each csv file because two learning models will be created. For example, you can create a model that learns welfare/qualification requirements + a model that learns tasks.

In step S110, the reason why the data processing unit 100 uses two learning models is to recommend companies that have similar criteria in one of 'welfare/qualification requirements' to the company entered by the user. If the ‘qualification requirements’ are similar but the actual job duties are different, the recommendation has no meaning to the user.

Employers attempt to find jobs in their field of expertise, and generally the types of jobs for any specialty are limited.

The data processing unit 100 performs learning on the tokenized data using the Doc2Vec model. At this time, the hyperparameter of the Doc2Vec model can be adjusted to select the value with the best accuracy.

The reason why Doc2Vec was used as a machine learning method at this time was because it was judged that Doc2Vec would be more advantageous for learning in that it is faster than the document embedding method using TensorFlow and can be learned by adjusting several hyperparameters.

In step S110, the data processing unit 100 may preferentially extract an index of companies that are more than 70% similar to the company entered by the user from the model that learned the task among the two learned models.

In step S110, the data processing unit 100 set the lower limit of job similarity to 70% because the amount of data used when training the job model (a model that learned the task column) was large (in the task column This is because, even though there was a 70% similarity, it was similar when comparing the actual data.

In step S110, the data processing unit 100 extracts similar companies using the 'welfare/qualification requirements' model for companies extracted from the job model and outputs them in order of highest similarity. For example, ultimately, companies with more than 70% similar jobs and similar 'welfare/qualification requirements' can be extracted.

The job matching method according to an embodiment of the present invention may include learning and calculating the correlation between companies satisfying the company conditions and company preferences according to at least one user specification (S120). .

In step S120, the learning unit 200 may learn and calculate the correlation between companies satisfying the company conditions and company preferences according to at least one user specification. The learning unit 200 can select and list companies that satisfy the company conditions. The learning unit 200 may rank companies that the user may prefer according to the correlation of company preferences according to the user's specifications from the list of companies that satisfy the company conditions.

In step S120, the learning unit 200 may match at least one user specification and company preference. The learning unit 200 may learn the correlation between the change in user specifications and the change in company preference by defining it as a probability. The learning unit 200 may calculate a correlation between the company conditions and company preferences according to at least one user specification according to the probability.

In step S120, the learning unit 200 may measure the correlation by defining the correlation between the change in the weather state attribute and the change in the state of the object as a probability.

In step S120, the learning unit 200 may calculate the influence of companies satisfying the company conditions and company preferences according to at least one user specification.

In step S120, the learning unit 200 may calculate the influence as a value that varies depending on how much influence a change in the user specification has on the company's preference using Equation 1 below.

[Equation 1]

Here, p(x) is the probability that a change in preference x will occur. As the probability of occurrence approaches 1, the influence approaches 0, requiring learning of a small number of cases, thus reducing the amount of computation. As the probability of occurrence approaches 0, the influence becomes infinitely greater, requiring learning of many cases, which may increase the amount of computation.

In step S120, the computational load in the learning unit 200 represents the expected value (average) of the influence. A large computational load means a large average influence, and the greater the uncertainty, the more difficult it is to classify, so the computational load is the highest. The amount of computation can be reduced by placing small items in the upper decision-making nodes. The computational load can be obtained by calculating Equation 2 below.

[Equation 2]

Here, S refers to the collection of all events that have already occurred, and c may refer to the number of events.

In step S120, the learning unit 200 can obtain the correlation by calculating Equation 3 below.

[Equation 3]

Here, C(S, P) is the correlation, E(S) is the analysis relevance value, S is the user specification, and P is the company preference. Here, P refers to a specific specification among user specifications, and it can be determined which specification has the lowest computational load (i.e., the largest amount of information acquisition) when classified.

The job matching method according to an embodiment of the present invention may include determining the ranking of companies that satisfy the company conditions according to the correlation (S130).

In step S130, the learning unit 200 may determine the ranking of the user specifications according to the correlation. The learning unit 200 may determine the ranking in descending order of correlation. The learning unit 200 may prioritize the user specifications with the low correlation in recommendation.

The user specifications may include at least one of school, major, gender, age, region, and MBTI.

In step S130, the learning unit 200 may determine a ranking including company preference according to a combination of a plurality of user specifications.

The job matching method according to an embodiment of the present invention may include presenting a list of companies to the user according to the ranking of the companies according to the correlation (S140).

In step S140, the visualization unit 300 may present a list of companies to the user according to the ranking of the companies according to the correlation.

In step S140, the visualization unit 300 may provide results according to the ranking of correlations between corporate preferences according to a plurality of user specifications. The visualization unit 300 may provide results according to the ranking of correlations between corporate preferences according to a plurality of combinations of the user specifications.

In step S140, the visualization unit 300 may visualize the data and provide it to the user so that it is easy to visually check which parts are similar when the user is recommended a similar company and checks the information in the recommended company announcement. . The visualization unit 300 can use the Wordcloud method to variously visualize key words that frequently appear in each company's announcement content. The visualization unit 300 displays similar companies on an additional web page using TSNE (a technique for reducing the dimension of high-dimensional vector data) on a two-dimensional coordinate plane, allowing users to explore similar companies in an intuitive and interesting way. Functions that can be used can also be provided.

In step S140, the visualization unit 300 can load and output the system previously developed up to the 'data embedding' stage on a web server using Flask. Data transmission between the Backend and FrontEnd is performed by sending data to the front-end as a parameter when the back-end calls the render_template function that shows the HTML file to the client. Data received from the back-end can be displayed in components placed on HTML using JavaScript and jQuery at the front-end. Data transfer from the front-end to the web server can be implemented by using the get method of the form tag, or by including data in the URL called by the front-end and then processing it at the back-end.

The web UI can be constructed intuitively and simply using various elements of bootstrap (ex. progressbar, radio button, table, card).

If the user needs to wait for the system to perform an operation, such as when the system is in the process of learning something new or creating visualization data, a progressbar can be used to let the user know what operation is in progress.

The composition of HTML elements can be designed for placement and size using CSS grammar in the style tag of <head>.

In the case of html files showing visualization information, new information must be provided rather than data cached in the web browser, so use Cache-Control to prevent caching and add a random number to the URL when calling a visualization image. By adding it and sending it to the server, the changed image can be provided to the user even if the image file of the same name is overwritten with other visualization information.

The structure of an example web page is as follows.

Start page: hello.html,

Job posting link input page: apply.html,

Loading page: index.html,

Analysis result page: result.html,

Company job posting details page: linktoimg.html,

Page notifying that there is no 'welfare' information: nobokri.html,

A page indicating that there are no similar companies: nolist.html;

Page showing visualization information: display_plot.html

The user enters the Job Planet job posting link in apply.html and then enters what criteria (benefits or qualifications) they would like to receive recommendations for similar companies.

In step S140, the visualization unit 300 extracts similar companies from the learned model according to the link and criteria entered by the user on the web server and provides them to the user in a table format through a table in result.html.

In step S140, the visualization unit 300 allows the user to click [View Details] for each company in result.html to receive WordCloud results that analyze keywords in job postings for the company and raw job posting content that has not been preprocessed. .

In step S140, if the visualization unit 300 commands a recommendation as welfare benefits without information on 'welfare' in the link entered by the user, the system re-directs the link to the user by showing nobokri.html. Prompts the user for input.

In step S140, if there is no company similar to the company in the link entered by the user (when there is no company similar to the company in both job duties and standards), the system displays nolist.html to indicate that there are no similar companies. Notifies the user and recommends other companies with similar 'jobs' although the 'input criteria' are not similar.

In step S140, the visualization unit 300 can move from result.html to display_plot.html, which shows visualization information.

In step S140, the visualization unit 300 draws a scatter plot on HTML using Plotly.js using the TSNE coordinate data received from the server in display_plot.

In step S140, the visualization unit 300 displays two plots: a scatter plot visualizing companies with similar 'user-selected criteria' to the user input company, and a scatter plot visualizing companies with similar 'jobs'. Shows the plot.

By clicking on the marker of each plot, the user is moved to linktoimg.html to receive job posting information from the company corresponding to that marker.

In step S140, the visualization unit 300 provides not only WordCloud results and raw job posting content in linktoimg.html, but also 'similarity' according to each criterion so that the user can directly check the reliability of the data recommended by the system. I do it.

Also, in linktoimg.html, depending on the marker the user clicks on the 'Visualization Results' page, if the red marker is clicked, it is notified that it is a job advertisement entered by the user, and if the blue marker is pressed, the job advertisement is notified that the job advertisement is a job advertisement entered by the user. They were asked to provide what criteria they analyzed as similar.

The embedding vector of each job posting is 200-dimensional high-dimensional data.

In step S140, in order to provide visualized information to the user, the visualization unit 300 reduces the data to two dimensions using a TSNE module and then visualizes it on a two-dimensional plane and provides it.

In step S140, the visualization unit 300 does not visualize information on all companies, but only visualizes and provides companies that are a certain distance away from the user input company (most similar companies).

In step S140, the visualization unit 300 operates when the distance between any two markers is very small among the visualized information, in order to prevent the phenomenon of appearing as one marker due to the very small difference in position between any two markers. , perform visualization by slightly changing the position of one of the two so that they do not overlap.

In step S140, the visualization unit 300 creates a WordCloud module using tokenized welfare/qualification data as a keyword and provides it.

At this time, if the user selects 'welfare' as the standard for determining similarity, the WordCloud result created with 'welfare' tokenized data is shown, and if the user selects 'qualification requirements', the WordCloud result created with 'qualification requirement' tokenized data is displayed. Shows the results.

So far, the present invention has been examined focusing on its preferred embodiments. A person skilled in the art to which the present invention pertains will understand that the present invention may be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered from an illustrative rather than a restrictive perspective. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the equivalent scope should be construed as being included in the present invention.

Claims (10)

A step of receiving input of at least one company condition and user specification desired by the user;
learning and calculating a correlation between companies satisfying the company conditions and company preferences according to at least one user specification;
determining the ranking of companies satisfying the company conditions according to the correlation; and
A job matching method comprising: presenting a list of companies to the user according to the ranking of companies according to the correlation. According to paragraph 1,
The correlation calculation step is,
Matching at least one user specification and company preference;
learning the correlation between the change in the user specifications and the change in the company's preference by defining it as a probability; and
A job matching method comprising: calculating a correlation between the company conditions and company preferences according to at least one user specification according to the probability. According to paragraph 2,
The correlation calculation step is,
A job search matching method, characterized in that the correlation is calculated by Equation 1 below.
[Equation 1]

(Here, C(S, P) is the correlation, E(S) is the analysis relevance value, S is the user specification, and P is the company preference.) According to paragraph 1,
The user specifications are:
A job matching method characterized by including at least one of school, major, gender, age, region, and MBTI. According to paragraph 4,
The step of determining the ranking of the above companies is,
Determine a ranking including company preference according to a combination of a plurality of user specifications,
The user presentation step is,
A job matching method characterized by providing results according to the above rankings. A data processing unit that receives at least one input of corporate conditions and user specifications desired by the user;
a learning unit that learns and calculates a correlation between companies that satisfy the company conditions and company preferences according to at least one user specification;
a ranking unit that determines the ranking of companies that satisfy the company conditions according to the correlation; and
A job matching system comprising a visualization unit that presents a list of companies to the user according to the ranking of companies based on the correlation. According to clause 6,
The learning department,
Match at least one user specification and corporate preference, define and learn the correlation between the change in the user specification and the change in the corporate preference as a probability, and learn the company according to the corporate condition and at least one user specification according to the probability. A job matching system characterized by calculating correlations between preferences. In clause 7,
The learning department,
A weather information analysis system based on driving vehicle information, characterized in that the correlation is calculated using Equation 1 below.
[Equation 1]

(Here, C(S, P) is the correlation, E(S) is the analysis relevance value, S is the user specification, and P is the company preference.) According to clause 8,
The user specifications are:
A job matching system characterized by including at least one of school, major, gender, age, region, and MBTI. According to clause 9,
The learning department,
Determine a ranking including company preference according to a combination of a plurality of user specifications,
The visualization unit,
A job matching system characterized by providing results according to the above rankings.