KR102473854B1 - Industrial occupation code classificational system and the method including the same - Google Patents

Abstract

The present invention relates to an industrial occupational code classification system and method thereof, which includes a text receiving unit for receiving a plurality of input texts, and code analysis for analyzing a matching code list by analyzing the received plurality of texts and providing matching codes. and a code transmission unit for transmitting a searched classification code and a keyword corresponding to the keyword to a user terminal through a network, wherein the code analysis unit matches the keyword of the text based on a dictionary to obtain a code list and a code list. A rule-based conversion module that provides reliability, a machine learning-based conversion module that converts/accumulates the text into an arbitrarily set feature vector and provides matching code lists and reliability, and converts the text to a random frequency for each code. By including a statistics-based conversion module that analyzes according to the code list and provides reliability, it provides the advantage of providing industrial and occupational codes with excellent accuracy and reliability.

Description

Industrial occupation code classification system and its method {INDUSTRIAL OCCUPATION CODE CLASSIFICATIONAL SYSTEM AND THE METHOD INCLUDING THE SAME}

The present invention relates to an industrial occupation code classification system and method.

Text classification studies have been conducted to provide various services. For example, research on emotion or emotion classification for text on the Internet has been conducted, and research on classifying service domains from text is also being conducted for the development of chatbots or conversation systems that have recently been in the limelight. These studies have been mainly based on rules or dictionaries in the past, and this method has the advantage that the engineering process for improving the performance of classification models is relatively easy, but it has the disadvantage that it becomes difficult to manage as the size of the rules or dictionaries increases. In order to overcome this problem, a statistical model was applied to a large amount of data to classify text, but the statistical model did not fully reflect the semantic and structural characteristics inherent in the text, and thus had limitations in performance. Machine learning models, including deep learning models that have recently been in the limelight, have achieved high performance by identifying semantic and structural features inherent in text, and are widely used in various fields as well as text data.

Overfitting or underfitting problems often occur when the complexity of the machine learning model and the complexity of the data are not balanced. In the case of industry and occupation code classification, the length of text input by the user is short, but the number of codes is very large, so a very large amount of data must exist to achieve a certain degree of classification accuracy. In other words, if existing machine learning-based research methods are applied as they are to the problem of industrial and occupational code classification, performance below expectations is bound to be obtained. Therefore, it is necessary to achieve a certain level of performance despite the lack of data by applying various methods in a complex manner.

In addition, the industrial and occupational code system changes, but no specific solution has been devised to solve this problem. In a situation where a relatively large amount of data is required due to the large number of codes, this code system change makes the problem more difficult.

Republic of Korea Patent Publication No. 10-2002-0004923 (2002.01.16. Publication date)

The present invention has been made to solve the above technical problem, and by applying a specific method to a vast industry and occupation code search, an industrial occupation code classification system that can provide results according to search input with greater accuracy, and its Its purpose is to provide a method.

In addition, the present invention provides an industrial occupational code classification system and method that can be easily utilized for industrial and occupational code classification by automatically converting the old system's industrial occupational code to the new system's industrial occupational code, for that purpose

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

In order to achieve the above object, the industrial occupation code classification system according to the present invention,

a text receiving unit for receiving a plurality of input texts;

a code analyzer configured to analyze a list of codes matched by analyzing the plurality of received texts and provide a matched code; and

A code transmission unit for transmitting the searched classification code and keyword corresponding to the keyword to a user terminal through a network;

The code analysis unit,

a rule-based conversion module that provides a code list and reliability by matching keywords of the text based on a dictionary;

A machine learning-based conversion module that converts/accumulates the text into an arbitrarily set feature vector and provides a matching code list and reliability therefrom; and

It may include a statistics-based conversion module that analyzes the text according to an arbitrary frequency for each code and provides a code list and reliability.

In one embodiment according to the present invention, the system includes a management server for storing and outputting data; and a user terminal communicating with the management server through a wired or wireless network.

In one embodiment according to the present invention, a code system conversion unit for analyzing the code list matched by the code transmission unit and converting it into a new code list; may be further included.

In one embodiment according to the present invention, the rule-based conversion module may be configured to set a code priority, search for a preset prohibited keyword among the input text, and designate it as a lower priority.

In one embodiment according to the present invention, the machine learning-based conversion module provides only a code list having a reliability greater than a predetermined threshold among a plurality of code lists matched from the text through the learned model. can

In one embodiment according to the present invention, the feature vector of the machine learning-based conversion module may include at least one of n-grams, the number of characters, special characters, and numbers.

In one embodiment according to the present invention, the model of the machine learning-based conversion module may include at least one of a decision tree, a support vector machine, a random forest, and deep learning.

In one embodiment according to the present invention, the arbitrary frequency of the statistics-based conversion module may be stored for each letter of the input text and displayed as a percentage (%) in order of high matching degree according to the frequency.

In one embodiment according to the present invention, at least one or more of the rule-based conversion module, the machine learning-based conversion module, and the statistics-based conversion module of the code analysis unit may be sequentially or arbitrarily operated according to the number of final matched code lists. .

On the other hand, the present invention provides a method for newly reclassifying industrial occupation codes using the industrial occupation code classification system,

a text receiving step of receiving a plurality of input texts;

Analyzing the plurality of received texts to analyze a matching code list,

A rule-based analysis process that matches keywords of the text based on a dictionary to provide a code list and reliability, converts/accumulates the text into an arbitrarily set feature vector, and provides a matching code list and reliability sequentially or randomly performing at least one of a machine learning-based analysis process and a statistical analysis process of analyzing the text according to an arbitrary frequency for each code and providing a code list and reliability; and

Comparing the matched code list with an existing code to determine whether there is a change, and if there is no change, providing a corresponding code list and reliability, or providing a new code list and reliability if a change is required; may include. .

In one embodiment according to the present invention, the rule-based analysis process of analyzing the code list, the machine learning analysis process, and the statistics-based analysis process may be repeatedly performed when the number of finally matched codes is less than the preset number of codes. have.

In one embodiment according to the present invention, the process of changing the code list includes the process of modifying the existing code, the process of modifying the definition of the existing code, the process of defining a new code, and the process of deleting the existing code. can do.

The industrial occupation code classification system and method according to the present invention have an effect of providing a result according to a search input with better accuracy by applying a specific method to a vast industry and occupation code search.

In addition, the industrial occupation code classification system and method according to the present invention automatically converts the industrial occupation codes of the old system into the industrial occupation codes of the new system based on the old system, so that it can be easily used for industrial and occupation code classification. have

1 is a schematic configuration diagram of an industrial occupation code classification system according to an embodiment of the present invention;
Figure 2 is an exemplary diagram by the industrial occupation code classification system according to an embodiment of the present invention,
3 to 5 are flowcharts showing an industrial occupation code classification process according to an embodiment of the present invention,
6 is a flowchart illustrating an industrial occupation code classification process according to an embodiment of the present invention.

Hereinafter, an embodiment of an industrial occupation code classification system and method according to the present invention will be described in detail with reference to the accompanying drawings. In adding reference numerals to components of each drawing, it should be noted that the same components have the same numerals as much as possible even if they are displayed on different drawings. In addition, in describing an embodiment of the present invention, if it is determined that a detailed description of a related known configuration or function hinders understanding of the embodiment of the present invention, the detailed description thereof will be omitted.

In describing the components of the embodiment of the present invention, terms such as first, second, A, B, (a), and (b) may be used. These terms are only used to distinguish the component from other components, and the nature, order, or order of the corresponding component is not limited by the term. In addition, unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the present application, they should not be interpreted in an ideal or excessively formal meaning. don't

Throughout the specification, when a part is said to 'include' or 'include' a certain component, it means that it may further include other components, not excluding other components unless otherwise stated. .

In general, in the prior art, there has been a problem in that resource consumption is large and accuracy is greatly reduced in the process of inspecting vast results searched in the process of finding codes for managing business fields, job types, and the like.

Accordingly, in the present invention, a code system with high accuracy and reliability is applied by applying a conversion method based on rules, machine learning, and statistics so as to have sufficient search results and high reliability beyond simply searching and matching the old code system. A system and method that can be provided can be provided.

Figure 1 is a schematic configuration diagram of an industrial occupation code classification system according to an embodiment of the present invention, Figure 2 is an exemplary diagram by the industrial occupation code classification system according to an embodiment of the present invention.

1 and 2, the industrial occupational code classification system 100 according to the present invention is a system that provides a code list and reliability by matching keywords derived from a plurality of input texts. A text receiving unit 150 that receives the text of the received text, a code analysis unit 100 that analyzes a matching code list by analyzing the plurality of received texts and provides a matching code, and the searched classification code and keyword corresponding to the keyword It may include a code transmission unit 160 for transmitting to the user terminal 200 through the network.

Here, the code analysis unit 100 is a component that matches and converts the old classification code 10 into the new classification code 20 and provides the code list and reliability by matching the keywords of the text based on a dictionary. A rule-based conversion module 110 that provides, a machine learning-based conversion module 120 that converts/accumulates the text into an arbitrarily set feature vector and provides a matching code list and reliability therefrom, and converts the text into each code It may be configured to include a statistics-based conversion module 130 that analyzes each code according to an arbitrary frequency and provides a code list and reliability. Also, the code analysis unit 100 may include a code system conversion module 140, which will be described below.

The text receiving unit 150 may receive a plurality of texts input through a wired or wireless network. Such text is provided to the code analysis unit 100, and the code analysis unit 100 may analyze, match, and convert a plurality of input texts in character units. The text that can be analyzed, matched, and converted in this way is processed through the rule-based conversion module 110, the machine learning-based conversion module 120, and the statistics-based conversion module 130 of the code analysis unit 100 described below. can be analyzed and transformed.

The rule-based conversion module 110 of the code analysis unit 100 can find and match an appropriate code from the old classification code 10 by checking conditions including text matching with respect to the text input through the text receiving unit 150. have. The rule-based conversion module 100 may perform keyword matching based on a dictionary, receives text input through the text receiving unit 150, checks whether such text exists, and provides two appropriately matched keywords. More than one code can be provided along with a confidence figure. In this case, priorities may be set between keywords for a plurality of codes matched through an already defined dictionary, or codes of all matched keywords may be provided. In a specific example, as shown in FIG. 2 , a matching code may be provided in the form of “0111” (reliability: 90%) for a keyword input as text “rice farming”. Of course, any number of matching codes may be provided based on the required number of promising rankings.

The aforementioned rule-based conversion module 100 may prioritize and provide specific keywords when a plurality of keywords are matched. For example, when “Hwawon-myeon office” is input to the text receiving unit 150, “Hwawon” is recognized as an A code and “Myeon office” is recognized as a B code, and A codes and A codes according to preset code priorities. Among the B codes, the B code can be given priority and output (B code: reliability (90%), A code: reliability (10%)).

Furthermore, the rule-based conversion module 110 of the code analysis unit 100 may search not only keyword matching of the text input from the text receiving unit 150 but also prohibited keywords that should not exist in units of keywords. For example, when “hop cultivation” is input to the text receiving unit 150, “hope” is recognized as an A code and “cultivation” is recognized as a B code, and the A code is deleted according to a preset prohibited keyword. and output (B code: reliability (100%)).

In addition to this example, the rule-based method of the rule-based transformation module 100 may check all possible cases with conditional statements (If ~, Then ~, Else ~) and provide an appropriate code list with reliability. The reliability value of each code may be arbitrarily calculated based on the priority of rules or the number of matched keywords. In addition, there may be cases where there is no matching through this rule-based method, and zero or more code results may be provided.

In addition, the machine learning-based conversion module 120 of the code analysis unit 100 expresses the text input through the text receiving unit 150 as an arbitrary feature vector and applies it to the machine learning model You can match the appropriate code. A vector of these arbitrary features can be arbitrarily defined by the model design of the system according to the present invention, and representative features include n-gram features, the number of letters, whether special characters or numbers appear, etc. can be included Here, N-gram features can be defined based on arbitrary units (eg letters, words, etc.). The input text is converted into such a feature vector, and a machine learning model can be learned based on the feature vector value.

Here, the machine learning model may include decision tree, support vector machine, random forest, deep learning model, and the like. A plurality of promising codes from the text input through the learned machine learning model may be provided together with reliability information. For these results, only some codes having higher reliability than a certain threshold may be extracted and provided. The machine learning-based method according to the machine learning-based conversion module 120 may generate zero or more code results.

In addition, the statistics-based conversion module 130 of the code analyzer 100 may analyze and convert codes based on statistics in arbitrary units (eg, letters).

When the text input through the text receiving unit 150 is statistically generated in character units, when N is the number of unique characters present in the training data, each code can be expressed as a frequency vector in N-dimensional space. For example, if the character "Nong" appears 10 times in the code "0111", it can be expressed in a format such as [??, 'Nong' = 10, ??].

For such learning data, after storing the frequency of arbitrary units of each letter for each code, it accumulates which code is more promising for each unit (eg, letter) for the text input through the text receiving unit 150. For example, if the text "Farming" is processed character by character, the code distribution for the word 'Farming' will be obtained as ["0111": 10%, "0112", 8%, ??], The code distribution for the letter 'four' is also obtained and summed up. In this way, the result is generated based on the code distribution summed up for each unit. Similarly, these statistics-based methods can also generate zero or more code results by applying arbitrary thresholds.

According to the present invention, the text analysis/conversion method of the above-described rule-based conversion module 110, the text analysis/conversion method of the machine learning-based conversion module 120, and the test analysis/conversion method of the statistics-based conversion module 130 are They can be applied in any order to provide analytical results. That is, this method can be readjusted through an arbitrary algorithm with respect to the results performed sequentially or extracted through each conversion module. Results may be changed by the ranking rebalancing algorithm.

For example, in the sequential execution method, the rule-based conversion module 110, the machine learning-based conversion module 120, and the statistics-based conversion module 130 may be operated in order, and in this case, the rule-based conversion module generated by the method You can prioritize results. If a sufficient number of code results are not generated even after each method of the rule-based conversion module 110, the machine learning-based conversion module 120, and the statistics-based conversion module 130 is performed, the reliability threshold value ( The sequential processing can be repeated until the output generated by reducing the threshold is sufficient.

In addition, in the method of performing the random order, the rule-based conversion module 110, the machine learning-based conversion module 120, and the statistics-based conversion module 130 may be randomly performed. The ranking of all results is re-adjusted through a random algorithm, and this random algorithm can be defined by a weight for each method or a priority rule between results. Similarly, if a sufficient number of code results are not generated even after each method of the rule-based transformation module 110, the machine learning-based transformation module 120, and the statistics-based transformation module 130 is performed, the reliability threshold value Sequential processing can be repeated until the output produced by reducing the threshold is sufficient.

Therefore, when proceeding through each analysis/conversion method according to the rule-based conversion module 110, the machine learning-based conversion module 120, and the statistics-based conversion module 130 as described above, the result according to the search input is viewed. It can be provided with excellent accuracy. In addition, the present invention may include a code system conversion module 140 of the code analysis unit 100. When the old classification code 10 is changed, the code system conversion module 140 can collect new data and convert it into the new classification code 20 .

Here, the operating process of the code system conversion module 140 is: First, a previously defined code number is changed, second, the definition of a previously defined code is changed, third, a new code is defined, and fourth, It may include the process of deleting the existing code.

Specifically, first, the existing code number change process may be performed by manually or automatically correcting the corresponding codes in the data of the old system code. Next, changes to the definition of existing codes can mainly include keywords or prohibited keywords. Therefore, first of all, specific data corresponding to these codes can be extracted and grouped, and different groups can be set according to whether or not they match one or more keywords. For example, if G(m) is the collection of data that matches one or more keywords and G(nm) is the data that does not match even a single keyword, G(nm) defines a new system code. It is possible to find and assign a new matching code by , and a data collection without a newly matching code can be referred to as G(u). These G(m) groups are matched with 'prohibited keywords' in the new code system, and those with even one banned keyword are found and changed to other codes by the new system code definition. At this time, data without a matching code in the new system code can be included in the Gu group.

Next, in the process of defining a new code, data can be randomly generated based on matching keywords by the above-described new code system, but forbidden keywords can be excluded. Finally, in the process of deleting the existing code, all data corresponding to the basic code may be included in the G(u) group. After the process is completed in this way, the data corresponding to the G(u) group can be manually searched for and tagged with appropriate keywords.

delete

delete

delete

delete

delete

delete

delete

delete

Steps of a method or algorithm described in relation to the embodiments presented herein may be implemented in the form of program instructions that can be executed by various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the medium may be those specially designed and configured for the present invention or those known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler. The hardware devices described above may be configured to act as one or more software modules to perform the operations of the present invention, and vice versa.

Although the present invention has been described in detail with reference to preferred embodiments, the present invention is not limited to the above embodiments, and the technical field to which the present invention belongs without departing from the gist of the present invention claimed in the following claims. Anyone skilled in the art will extend the technical spirit of the present invention to the extent that various variations or modifications are possible.

10: old classification code
20: new classification code
100: code analysis unit
110: rule-based change module
120: machine learning-based conversion module
130: statistics-based conversion module
140: code system conversion module
150: text receiving unit
160: code transmission unit
200: user terminal

Claims (12)

a text receiving unit for receiving a plurality of texts input through a wired or wireless network;
a code analyzer configured to analyze a list of codes matched by analyzing the plurality of received texts and provide a matched code; and
A code transmission unit for transmitting the searched classification code and keyword corresponding to the keyword to a user terminal through a network;
The code analysis unit,
a rule-based conversion module that matches keywords of the text based on a dictionary and provides a preset code list stored in a server;
A machine learning-based conversion module that converts/accumulates the text from the server into at least one feature vector selected from randomly set characters, words, the number of characters, special characters, and numbers, and provides a list of codes matched therewith; and
A statistics-based conversion module for analyzing the text according to an arbitrary frequency for each code and providing a list of matching codes;
The machine learning-based conversion module,
Provides only a code list having a result matching a keyword derived from a plurality of texts higher than a predetermined threshold among a plurality of code lists matched from the text through the learned model,
The model of the machine learning-based conversion module includes at least one of a decision tree, a support vector machine, a random forest, and deep learning,
The rule-based conversion module,
It is configured to set a code priority, search for a preset prohibited keyword among the input text, and designate it as a lower priority;
The rule-based conversion module, the machine learning-based conversion module, and the statistics-based conversion module of the code analysis unit are sequentially or randomly operated at least one according to the number of final matched code lists.
delete According to claim 1,
Industrial occupation code classification system further comprising: a code system conversion unit for analyzing the code list matched by the code transmission unit and converting it into a new code list.
delete delete delete delete According to claim 1,
The random frequency of the statistics-based conversion module is stored for each letter of the input text and displayed as a percentage (%) in the order of high matching according to the frequency.
delete delete delete delete

Images