KR20210050155A - Document screening system using artificial intelligence - Google Patents

Abstract

The present invention relates to an application document guide method using artificial intelligence. The method includes a document screening server: dividing a document created by an applicant into sentences; parsing each sentence of the document; acquiring document context information based on the parsing; determining the degree of importance of each sentence using the context information and information on the applied job of the applicant; and learning the sentence-specific degree of importance corrected by an evaluator in the document with a determined degree of importance. According to the present invention, a self-introduction sentence context can be understood, an important sentence can be provided with highlighting, non-highlighted sentences regarded by an evaluator as important sentences can be selected for feedback to a document screening system, and the evaluator's opinion can be reflected in the subsequent selection of important self-introduction sentences. Accordingly, the same evaluation criteria can be applied to every applicant.

Description

Document screening system using artificial intelligence {DOCUMENT SCREENING SYSTEM USING ARTIFICIAL INTELLIGENCE}

The present invention relates to a document screening system using artificial intelligence, and by using natural language processing technology to determine the context of an application written by an applicant and displaying important words and sentences on the application, the evaluator's decision on the application is quickly performed. It relates to a document screening system that can minimize the cost of manpower incurred to evaluate a large number of applications.

In general, for regular recruitment of new employees, a small number of practitioners/evaluators spend a lot of time reading and scoring hundreds of application documents in each company. Accordingly, in order to increase the efficiency of the evaluator's work, a technology for analyzing supporting documents using artificial intelligence (AI) has recently been launched.

Representatively, by analyzing the applicant's application documents using language processing analysis technology, there are methods such as whether the content is suitable for the company, whether there is a possibility of plagiarism, and whether the company's competency keywords exist, but the sentences plagiarized in the applicant's application documents. It is just a technology that checks whether or not there is a specific word for each company.

In other words, it can be said that the most accurate method is for the evaluator to read the application documents and grasp the context in order to determine whether the applicant is the right person for the company through the application documents. However, since the same evaluator cannot review each time it is hired, there is a problem that the evaluation score is inconsistent.

Therefore, it is required to develop a technology capable of providing a guide for document evaluation so that the evaluator can quickly read the application documents and make a decision, and the present invention relates to this.

Korean Patent Publication No. 10-2016-0137135 (2015.05.22.)

The technical problem to be solved by the present invention is to provide a guide for evaluating the self-introduction by understanding the context of the sentence of the supporting document (hereinafter, self-introduction) and highlighting the area judged as an important sentence. do.

Another technical problem to be solved by the present invention is to provide a guide for another self-introduction by reflecting the evaluator's feedback on the self-introduction letter (e.g., sentences that the evaluator considered important in addition to the highlighted sentence). The purpose of this is to provide a document screening system that can reflect this.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems that are not mentioned will be clearly understood by those skilled in the art from the following description.

A method for guiding a supporting document using artificial intelligence according to an embodiment of the present invention includes the steps of separating a document written by an applicant into sentences, and the document selection server analyzes the syntax of each of a plurality of sentences constituting the document. Step of, the document selection server obtaining context information of the document based on the analyzed syntax, the document selection server using the job information and the context information applied by the applicant Determining the importance of each of the plurality of sentences, and learning, by the document selection server, the importance of each sentence corrected by the evaluator in the document for which the importance is determined.

According to an embodiment, before the step of separating into sentences, the step of vectorizing words in the document or words in the previously stored document using a morpheme analysis, an entity name analysis, or a semantic role analysis method is further performed. Can include.

According to an embodiment, the sentences in the pre-stored document are composed of consecutive phrases, and words representing the core of each of the consecutive phrases are combined and extracted in the form of a first characteristic vector, and the first characteristic vector is obtained. It may further include the step of generating a word-vector transformation model based on.

According to an embodiment, before the step of separating into sentences, the pre-stored document is divided into sentences, context information of consecutive sentences from each sentence is extracted in the form of a second feature vector, and the second feature vector It may further include generating an importance prediction model based on.

According to an embodiment, in the generating of the importance prediction model, the importance of each sentence in the previously stored document is determined based on the degree of consensus of the evaluators on the same sentence or the document screening evaluation score of a previously stored document. And reflecting the determined importance of each sentence to the importance prediction model.

According to an embodiment, the information on the job includes a core element designated for each job, and determining the importance of each of the plurality of sentences includes relevance of each of the plurality of sentences with the core element and the context information. And assigning an importance score according to the assigned importance score, and designating a mark to be emphasized on a corresponding sentence in the document so that the evaluator can recognize it in each of the plurality of sentences according to the assigned importance score.

According to an embodiment, the vectorizing may be a step of classifying the pre-stored document according to a job standard, extracting words with an appearance frequency greater than or equal to a reference value in the classified document, and vectorizing it to have a similar vector value. .

A document screening server according to another embodiment of the present invention includes a sentence separation unit for separating a document written by an applicant into sentence units, a word-vector conversion unit for vectorizing words in the document, and a plurality of sentences constituting the document. A syntax analysis unit that analyzes the syntax of the sentence, a context acquisition unit that obtains context information of the document based on the analyzed syntax, and the job information and the context information written by the applicant are used for each of the plurality of sentences. It includes a sentence importance determining unit that determines the importance and learns the importance of each sentence corrected by the evaluator within the document in which the importance is determined.

According to an embodiment, the word-vector conversion unit may vectorize a word in the document or a word in a pre-stored document by using a morpheme analysis, an entity name analysis, or a semantic role analysis method.

According to an embodiment, the sentences in the pre-stored document are composed of consecutive phrases, and the syntax analysis unit combines words representing the core of each of the consecutive phrases and extracts them in the form of a first characteristic vector, and the A word-vector transformation model may be generated based on the first feature vector.

According to an embodiment, the context acquisition unit divides the pre-stored document into sentences, extracts context information of consecutive phrases from each sentence in the form of a second feature vector, and predicts importance based on the second feature vector. You can create a model.

According to an embodiment, the sentence importance determining unit determines the importance of each sentence within the previously stored document based on the degree of agreement of the evaluators' opinions on the same sentence or the document screening evaluation score of a previously stored document, and The importance of each sentence may be reflected in the importance prediction model.

According to an embodiment, the information on the job includes a core element designated for each job, and the sentence importance determining unit assigns an importance score to each of the plurality of sentences according to a correlation between the core element and the context information. , In accordance with the assigned importance score, a mark to be emphasized on a corresponding sentence in the document may be designated so that the evaluator can recognize in each of the plurality of sentences.

According to an embodiment, the word-vector converter may classify the pre-stored document on a job basis, extract words with an appearance frequency greater than or equal to a reference value in the classified document, and vectorize it to have a similar vector value.

According to the present invention, by understanding the sentence context of the self-introduction letter, highlighting the important sentences, and providing it to the evaluator, the time it takes for the evaluator to read the self-introduction and complete the judgment can be shortened and manpower costs can be reduced. I can.

In addition, since the document screening system provides only the highlighting of important sentences in the self-introduction letter, and the evaluator who decides pass/fail for this is used only as a reference, hiring can proceed more fairly and efficiently.

In addition, in addition to the highlighted sentences, the evaluator selects the sentences that the evaluator thinks as important sentences and feeds them back to the document transfer system. After that, the evaluator's opinion can be reflected when selecting the important sentences in the self-introduction letter, and it is reflected in the learning model. The same evaluation criteria can be applied to applicants.

The effects of the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a diagram schematically showing the configuration of a document screening system according to a first embodiment of the present invention.
2 is a diagram schematically showing the configuration of a document screening system according to a second embodiment of the present invention.
3 is a diagram schematically showing the configuration of a document screening server according to a second embodiment of the present invention.
4 is a diagram for explaining a function performed by each component in the configuration of the document screening system according to the second embodiment of the present invention.
FIG. 5 is a diagram illustrating a flow of a method in which the document screening server 200 learns important words and infers and guides important sentences from a supporting document according to a second embodiment of the present invention.
6A and 6B are views exemplarily showing a document that has been guided by a document selection server according to a second embodiment of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments to be posted below, but may be implemented in various different forms, and only these embodiments make the posting of the present invention complete, and common knowledge in the technical field to which the present invention pertains. It is provided to completely inform the scope of the invention to the possessor, and the invention is only defined by the scope of the claims. The same reference numerals refer to the same elements throughout the specification.

Unless otherwise defined, all terms (including technical and scientific terms) used in the present specification may be used with meanings that can be commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, terms defined in a commonly used dictionary are not interpreted ideally or excessively unless explicitly defined specifically. The terms used in this specification are for describing exemplary embodiments and are not intended to limit the present invention. In this specification, the singular form also includes the plural form unless specifically stated in the phrase.

As used herein, "comprises" and/or "comprising" refers to the presence or addition of one or more other components, steps and/or actions to the referenced components, steps and/or actions. Do not rule out.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically showing the configuration of a document screening system 1000 according to a first embodiment of the present invention, and FIG. 2 is a diagram schematically showing the configuration of a document screening system according to a second embodiment of the present invention .

Referring to FIG. 1, the document selection system 1000 may include a legacy system and an artificial intelligence system. Legacy System is an existing system used by each company for hiring new employees, and it can receive and store various application documents from applicants, and receive evaluation of the evaluator for the application documents through the evaluator terminal 300. I can.

The AI System is a system that can provide a guide for the evaluator to easily read and review hundreds of supporting documents, and is a learning server 10, a service server 20, and a document selection model ( 30).

Specifically, the learning server 10 1) can receive the results of the evaluation of the existing support documents from the recruitment server 100, and by learning this, 2) create a document screening model 30 to provide a guide for supporting documents, and Can be saved.

The service server 20 can 3) reflect the document selection model 30 generated by the learning server 10 to the artificial intelligence engine, and 4) when the supporting documents are provided from the recruitment server 100, the artificial intelligence engine is By utilizing 5) a guide for the application document may be generated (evaluation result) and provided to the recruitment server 100.

On the other hand, according to an embodiment of the present invention, one document screening server 200 may perform the functions of the learning server 10 and the service server 20, and the configuration including the document screening server 200 is illustrated in FIG. May be equal to 2.

So far, the entire configuration of the document screening system 1000 according to the first and second embodiments of the present invention has been examined. Hereinafter, a guide to the supporting documents so that the evaluator can read the application documents and make a decision easily. It will be described in detail with respect to the method of providing.

3 is a diagram schematically showing the configuration of a document screening server 200 according to the second embodiment of the present invention, and FIG. 4 is a diagram showing the configuration of the document screening system according to the second embodiment of the present invention. It is a diagram for explaining the function to be performed.

3 and 4, the document screening server 200 includes a sentence management unit 210, a word-vector conversion unit 220, a syntax analysis unit 230, a context acquisition unit 240, and a sentence importance determining unit ( 250) and the memory 260 may be included, and other additional components for achieving the object of the present invention may also be included.

Meanwhile, the sentence management unit 210, the word-vector conversion unit 220, the syntax analysis unit 230, the context acquisition unit 240, the sentence importance determining unit 250, and the memory 260 are an embodiment of the present invention. It is only a temporary or virtual configuration classified based on the function of the document screening server 200 according to, and the function performed by one configuration can be performed by another configuration, and one configuration can perform all the functions of the entire configuration. Of course it can be done. For example, one processor (not shown) includes a sentence management unit 210, a word-vector conversion unit 220, a syntax analysis unit 230, a context acquisition unit 240, a sentence importance determining unit 250, and a memory. All of the functions of 260 may be performed.

The sentence management unit 210 may separate the application documents (hereinafter referred to as'documents') prepared by the applicant into sentences and refine the contents of the document into a form capable of processing data. However, the sentence management unit 210 may check the items existing in the document before dividing all sentences in the document into individual sentences, and group sentences of the same category for each item. For example, the sentence management unit 210 may divide the applicant's document into a plurality of sentences as shown in [Table 1].

Before separation I think the experience of the POC project that developed the image recognition system is the most meaningful. The most important thing to keep your project running smoothly is to understand your requirements accurately. Together with the team members, we analyzed the solution manual, visited the developer, and tried to improve understanding through questions. We have developed a high-quality system through a multifaceted mindset and efficient logic design. After separation Sentence 1) I think the experience of the POC project that developed the image recognition system is the most meaningful.
Sentence 2) The most important thing for smooth project progress is to accurately understand user requirements.
Sentence 3) We analyzed the solution manual with the team members, visited the developer, and tried to improve understanding through questions.
Sentence 4) We have developed a high-quality system through multifaceted thinking and efficient logic design.

The word-vector conversion unit 220 may convert words in a vast amount of previously stored documents (documents that have been evaluated by evaluators) into vectors through a word-embedding model (eg, Doc2Vec). have. For example, the word-vector conversion unit 220 may vectorize each word using a morpheme analysis, an entity name analysis, or a semantic role method. In the case of morpheme analysis, only real morphemes having meaning Can be vectorized.

In addition, the word-vector conversion unit 220 may extract words with a frequency of occurrence in the document greater than or equal to a reference value and vectorize them to have a similar vector value, which may be used to learn a word-vector conversion model later. . Here, the reference value of the frequency of appearance may be calculated by calculating the frequency of appearance of each word relative to the total number of sentences.

Meanwhile, the word-vector conversion unit 220 may vectorize the job information and each word written by the applicant based on the feature vectors extracted by the syntax analysis unit 230 and the context acquisition unit 240 to have similar values. have.

The syntax analysis unit 230 may analyze the syntax of each of a plurality of sentences constituting a document. Specifically, the syntax analysis unit 230 may separate one sentence based on a phrase and a clause, or designate the number of words, and treat it as a plurality of phrases to separate them. Here, phrase analysis is a technique for separating and analyzing a long sentence into several phrases or clauses, and various natural language processing algorithms and machine learning algorithms may be used.

That is, the syntax analysis unit 230 may preferentially perform a process of learning important words from a document that has been evaluated previously stored in the memory 260 in order to select important words from a document written by the applicant. To this end, the syntax analysis unit 230 may extract k word combinations that influence the determination of the importance of a sentence from n word vectors in the form of a feature vector using a convolution neural network (CNN) method.

In other words, the syntax analysis unit 230 may extract a word combination (key phrase) representing the core of each sentence from a previously stored document in the form of a first feature vector, and a word-vector conversion model based on the first feature vector Can be created.

Meanwhile, the importance prediction model may be continuously updated by the evaluator during the document evaluation process.

In this way, the syntax analysis unit 230 may generate one core phrase by combining important words within a sentence. For example, a core phrase such as [Table 2] is included in each of the sentences separated in [Table 1]. Can be created.

Sentence 1 I think the experience of the POC project that developed the image recognition system is the most meaningful. Key phrase Image recognition system Sentence 2 The most important thing to keep your project running smoothly is to understand your requirements accurately. Key phrase Understanding user requirements Sentence 3 Together with the team members, we analyzed the solution manual, visited the developer, and tried to improve understanding through questions. Key phrase Solution Manual Analysis Sentence 4 We have developed a high-quality system through a multifaceted mindset and efficient logic design. Key phrase Efficient logic design

The context acquisition unit 240 may acquire context information of a document based on the syntax analyzed by the syntax analysis unit 230, and for this purpose, the context acquisition unit 240 is a recurrent neural network model. RNN) can be used. More specifically, the context acquisition unit 240 may analyze the context by combining the key phrases in order. Based on the first key phrase, the context of the second key phrase is derived, and the conclusion of the context is presented through the last key phrase. I can.

For example, when the order of key phrases is'Image Recognition System'-'Understanding User Requirements'-'Solution Manual Analysis'-'Efficient Logic Design', the context acquisition unit 240 is used as'system operation job suitability context'. Context information can be obtained.

In addition, the context acquisition unit 240 may extract the continuous phrase (context information) in the form of a second feature vector and generate a word-vector conversion model based on the second feature vector, and the word-vector conversion model It can be used in the process of converting a word in a newly received document into a word vector.

The sentence importance determining unit 250 may predict and determine the importance of each of a plurality of sentences by using job information and context information of a document written by the applicant. Here, the job information may include key elements (eg, challenge, positivity, global competency) designated for each job. Specifically, the sentence importance determining unit 250 may assign an importance score to each of the plurality of sentences according to the association between the core element and context information, and may designate a different display to each of the plurality of sentences according to the assigned importance score. That is, the score assigned to each sentence according to the relevance may be an indicator of the importance.

In other words, the sentence importance determining unit 250 1) assigns an importance score according to the presence or absence of a word related to a core element in each of a plurality of sentences, and 2) a core sentence representing context information derived through the plurality of sentences. By assigning an importance score to each sentence, the importance of each sentence can be determined comprehensively.

Referring to [Table 2], context information called'system operation job suitability context' was created by combining key phrases, and when the keyword of the job applied by the applicant is'system development', the sentence importance determining unit 250 each The importance of the sentence of can be determined as shown in [Table 3]. For reference, when the importance is expressed in a score format, the meaning of the numbers can be written, such as 2: important sentences, 1: correlation sentences, and 0: others.

Sentence example importance I think the experience of the POC project that developed the image recognition system is the most meaningful. 2 The most important thing to keep your project running smoothly is to understand your requirements accurately. 0 Along with the seniors in the PM role, we thoroughly analyzed the requirements definition and distributed tasks to implement the functions. 0 However, interns were frustrated by the lack of understanding of in-house solutions and the unfamiliar development environment. 0 Together with the team members, we analyzed the solution manual, visited the developer, and tried to improve understanding through questions. One We have developed a high-quality system through a multifaceted mindset and efficient logic design. 0

As shown in [Table 3], the sentence importance determining unit 250 can give a visual effect so that the evaluator can recognize the importance score assigned to each sentence in the document, and display highlighting of different colors in addition to the importance score. You can designate a different display for each sentence, such as in bold.

In addition, the sentence importance determining unit 250 can give a visual effect to all sentences having an importance score higher than the reference value, and the reference value here is set to a fixed value by the administrator, or the entire sentence in the document It can be set as the upper n% (n is a natural number greater than or equal to 1).

In addition, the sentence importance determining unit 250 may determine the importance of each sentence based on the degree of consensus of the evaluators on the same sentence in the previously stored document or the document screening evaluation score of the previously stored document. Can be reflected in the importance prediction model generated by

The memory 260 may store the application document received by the recruitment server 100 and the application document data reviewed by the evaluator, and store the document selection model 30 learned based on this.

So far, the configuration of the document screening server 200 according to an embodiment of the present invention has been described. According to the present invention, the document screening server 200 may analyze the document created by the applicant in detail by sentence and word and grasp the context of the document, thereby highlighting the sentence corresponding to the keyword set for each company. In addition, as a result of measuring the document review time of the applicant using the document screening server 200, the time required to evaluate one document on average is significantly reduced compared to the previous one, so that the review efficiency as well as the manpower cost can be reduced. In addition, as the amount of applications that the document screening server 200 learns continuously increases, the time required to evaluate one document may gradually decrease.

Hereinafter, a method of guiding the application documents using the above-described document selection server 200 will be described. For reference, guides designate specific marks on important sentences in the document so that the evaluator can read the application document more quickly.

FIG. 5 is a diagram illustrating a flow of a method in which the document screening server 200 learns important words and infers and guides important sentences from a supporting document according to a second embodiment of the present invention.

<Learning Department>

Referring to FIG. 5, the document screening server 200 learns the context information of documents that have been evaluated and pre-stored documents so that words having similar meanings have similar vector values. A vector transformation model for grasping is generated (S101). Specifically, the document screening server 200 may classify documents for which evaluation has been completed according to the similarity of the context, and extract a word corresponding to the context and a corresponding document into a word vector and a document vector having similar values, and based on this You can create a word-vector transformation model. For example, in the document screening server 200, the classified context information is'customer marketing job suitability context' and'customer service job suitability context', and words corresponding to the context information'request'-'understanding', ' It is possible to convert the words'needs'-'knowledge' into a vector of similar values. In other words, job documents such as'marketing' and'customer service' in the learning process can be learned by taking an average together with word vectors.The document vector has the same value as the word vector, so that the relative meaning between documents can be grasped. This is possible. In addition, in order to perform the process of converting such words into vectors, only real morphemes having meanings may be vectorized by morpheme analysis of words and documents.

After step S101, the document selection server 200 learns the importance of documents for which the evaluation has been completed, and generates an importance prediction model for predicting the importance (S102). Here, the evaluation-completed document is a document including a sentence that the evaluator regards as an important sentence and has been separately marked, or a document in which an important sentence has been designated by the document selection server 200 and the evaluator, and may include keywords for each job. In addition, importance is an index that quantitatively indicates to what extent each of the plurality of sentences constituting the document is suitable for the job applied by the applicant.For example, according to the setting of the evaluator, 0 is the most inappropriate sentence and 5 is the most suitable sentence. Can be tagged as. Here, as for the score for the sentence, various score ranges, which are not limited thereto, may be set.

Specifically, the document screening server 200 combines words representing the core of each phrase in the sentence for which the score for each importance is set (generates the core phrase), extracts it in the form of a first characteristic vector, and extracts context information of consecutive phrases from the sentence By extracting in the form of a second feature vector, an importance prediction model can be generated based on this, and the importance score by the evaluator is reflected, and the model can be continuously updated by the evaluator during the later document evaluation process of the applicant. .

For example, if the document screening server 200 reads the mark emphasized as important and it is determined that it is wrong, the evaluator can cancel the highlighting, and if it is determined that the part that the document screening server 200 turned over as not important is wrong, The evaluator may also select important sentences. In this way, since feedback from the evaluator is continuously provided and reflected in the learning data, it is possible to increase the fairness of the evaluation process.

As described above, the document screening server 200 converts important words in important sentences into word vectors using a word-embedding model, and then uses neural network models (CNN, RNN) to determine according to characteristics. By converting into a 1 feature vector and a second feature vector, a model can be generated to grasp the context of the document created by the applicant and assign importance.

So far, the document screening server 200 has described a method of learning the importance of the sentence for which the evaluation has been completed, and generating and learning a model for grasping the context. According to the present invention, since the opinion of each evaluator can be reflected in the learning model, the same evaluation criteria can be applied to all applicants.

<Inference Department>

The document screening server 200 separates the document created by the applicant into sentences (S110). This step is a step of refining the contents of the document into a form that can be processed by the document screening server 200, and all sentences in the document may be divided into individual sentences. However, before dividing into sentences, you can check the items that exist in the document, and group sentences of the same category for each item.

After step S110, the document selection server 200 analyzes the syntax of each of a plurality of sentences constituting the document (S120). That is, the document screening server 200 may perform syntax analysis after dividing the document into word groups that are smaller units in paragraph-sentence.

Accordingly, the document screening server 200 may divide each of the plurality of sentences into words or morphemes, select important words from the divided word groups, and generate the core phrases of the corresponding sentences. In this case, important words may be selected using the learned importance prediction model.

More specifically, the document screening server 200 extracts a combination of important words in the form of a feature vector and inputs the combination of important words into an importance prediction model to select important words. In addition, since the document screening server 200 inputs a vector value of an importance prediction model, a Euclidean distance similarity, a cosine similarity method, or the like may be used to measure the similarity between vectors.

After step S120, the document screening server 200 acquires context information of the document based on the analyzed syntax (S130). In order to obtain context information, the document selection server 200 may use a recurrent neural network (RNN) model, and may analyze the context of a document by combining key phrases in order.

For example, if the order of key phrases is'Image Recognition System'-'Understanding User Requirements'-'Solution Manual Analysis'-'Efficient Logic Design', the document screening server 200 is used as'system operation job suitability context'. Context information can be obtained.

After step S130, the document screening server 200 determines the importance of each of the plurality of sentences by using the job information and the context information of the document applied by the applicant (S140). Here, the job information may include key elements (eg, challenge, affirmation) designated for each job. Specifically, the document screening server 200 may assign an importance score to each of the plurality of sentences according to the association between the core element and context information, and may designate a different display for each of the plurality of sentences according to the assigned importance score. . That is, the score assigned to each sentence according to the relevance may be an indicator of importance.

In other words, 1) an importance score is assigned according to the presence or absence of a word related to a key element in each of the plurality of sentences, and 2) an importance score is assigned to the key sentence representing the context information derived from the plurality of sentences. You can determine the importance of each.

After step S140, the document selection server 200 learns the importance of each sentence corrected by the evaluator in the document whose importance is determined (S150). For example, the document screening server 200 may receive another sentence from the evaluator in addition to the sentence highlighted as an important sentence, and cancel the sentence highlighted as an important sentence. This can soon be fed back by the Ministry of Education as a material for learning the application documents.

Meanwhile, the document screening server 200 may provide a visual effect so that the evaluator can recognize the score assigned to each sentence in the document.

In this regard, FIGS. 6A and 6B are diagrams showing exemplary documents that have been guided by the document selection server 200 according to the second embodiment of the present invention, and referring to FIGS. 6A and 6B, You can see that the sentences are shaded so that the evaluator can recognize them at a glance. In addition, key elements (motivation for application, global competency) designated for each job are displayed so that the evaluator can read it and easily evaluate the document.

On the other hand, the sentence may be marked with a hatched mark and an importance score, or may be displayed in bold, underlined, or highlighted in different colors depending on the importance.

So far, a method for guiding important sentences in a document by the document screening server 200 according to an embodiment of the present invention has been described for the convenience of the evaluator. According to the present invention, since the document is not numerically reviewed, such as whether a document is plagiarized, whether or not there is a keyword for each company, and the number of keywords, it is necessary to understand the context and review whether the document is written in accordance with the The evaluator who decides to fail can make decisions more easily.

On the other hand, the present invention can also be implemented as a computer-readable code on a computer-readable recording medium. Computer-readable recording media include all storage media such as magnetic storage media and optical reading media. In addition, it is possible to record the data format of the message used in the present invention on a recording medium.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, those of ordinary skill in the art to which the present invention pertains can be implemented in other specific forms without changing the technical spirit or essential features. You will be able to understand. Therefore, it should be understood that the embodiments described above are illustrative and non-limiting in all respects.

1000: document screening system
100: Recruitment Server
200: document screening server
10: learning server
20: service server
30: document screening model
210: sentence separator
220: word-vector conversion unit
230: parsing unit
240: context acquisition unit
250: sentence importance determining unit
260: memory
300: evaluator terminal

Claims (14)

Separating, by the document selection server, the document created by the applicant in sentence units;
Analyzing, by the document selection server, a phrase of each of a plurality of sentences constituting the document;
Obtaining, by the document selection server, context information of the document based on the analyzed syntax;
Determining, by the document screening server, the importance of each of the plurality of sentences using information on the job applied by the applicant and the context information; And
Learning, by the document selection server, the importance of each sentence corrected by the evaluator in the document for which the importance is determined;
How to guide the supporting documents using artificial intelligence, including. The method of claim 1,
Before the step of separating into the sentence units,
Vectorizing a word in the document or a word in a pre-stored document using a morpheme analysis, an entity name analysis, or a semantic role analysis method;
How to guide the supporting documents using artificial intelligence, which further includes. The method of claim 2,
Before the step of separating into the sentence units,
Sentences in the pre-stored document consist of consecutive phrases,
Combining words representing the core of each of the consecutive phrases, extracting them into a first feature vector form, and generating a word-vector transformation model based on the first feature vector;
How to guide the supporting documents using artificial intelligence, which further includes. The method of claim 2,
Before the step of separating into the sentence units,
Dividing the pre-stored document into sentences, extracting context information of consecutive phrases from each sentence in the form of a second feature vector, and generating an importance prediction model based on the second feature vector;
How to guide the supporting documents using artificial intelligence that further includes. The method of claim 4,
Generating the importance prediction model,
Determining the importance of each sentence in the previously stored document based on the degree of agreement of the evaluator's opinion on the same sentence or the document screening evaluation score of the previously stored document, and reflecting the determined importance of each sentence in the importance prediction model ;
How to guide the supporting documents using artificial intelligence, which further includes. The method of claim 1,
The information of the above job is,
Includes key elements specified for each job,
Determining the importance of each of the plurality of sentences,
An importance score is assigned to each of the plurality of sentences according to the correlation between the core element and the context information, and the corresponding sentence is emphasized in the document so that the evaluator can recognize it in each of the plurality of sentences according to the assigned importance score. Designating an indication to be performed;
How to guide the supporting documents using artificial intelligence, which further includes. The method of claim 2,
The vectorizing step,
The step of classifying the pre-stored document based on job criteria, extracting words with a frequency of occurrence of a reference value or more in the classified document, and vectorizing them to have a similar vector value,
How to guide application documents using artificial intelligence. A sentence separating unit for separating the document written by the applicant into sentence units;
A word-vector conversion unit for vectorizing words in the document;
A syntax analysis unit that analyzes a syntax of each of a plurality of sentences constituting the document;
A context acquisition unit that acquires context information of the document based on the analyzed syntax; And
A sentence importance determining unit for determining the importance of each of the plurality of sentences using the job information and the context information written by the applicant, and learning the importance of each sentence corrected by the evaluator in the document in which the importance is determined;
Document screening server comprising a. The method of claim 8,
The word-vector conversion unit,
Vectorizing a word in the document or a word in a pre-stored document using a morpheme analysis, entity name analysis, or semantic role analysis method,
Document screening server. The method of claim 9,
Sentences in the pre-stored document consist of consecutive phrases,
The parsing unit,
Combining words representing the core of each of the consecutive phrases, extracting them in the form of a first feature vector, and generating a word-vector conversion model based on the first feature vector,
Document screening server. The method of claim 9,
The context acquisition unit divides the pre-stored document into sentences, extracts context information of consecutive phrases from each sentence in the form of a second feature vector, and generates an importance prediction model based on the second feature vector,
Document screening server. The method of claim 11,
The sentence importance determining unit,
Based on the degree of agreement of the evaluator's opinion on the same sentence or the document screening evaluation score of a previously stored document, the importance of each sentence is determined in the previously stored document, and the determined importance of each sentence is reflected in the importance prediction model,
Document screening server. The method of claim 8,
The information of the above job is,
Includes key elements specified for each job,
The sentence importance determining unit,
An importance score is assigned to each of the plurality of sentences according to the correlation between the core element and the context information, and the corresponding sentence is emphasized in the document so that the evaluator can recognize it in each of the plurality of sentences according to the assigned importance score. To designate a sign to do,
Document screening server. The method of claim 9,
The word-vector conversion unit,
Classifying the pre-stored document based on a job standard, extracting words with a frequency of occurrence greater than or equal to a reference value in the classified document, and vectorizing it to have a similar vector value,
Document screening server.

Images