KR100837797B1 - Method for automatic construction of acronym dictionary based on acronym type, Recording medium thereof and Apparatus for automatic construction of acronym dictionary based on acronym type - Google Patents

Abstract

Disclosed are a method for automatically building an abbreviation dictionary that takes into account an abbreviation generation type, and an apparatus for automatically building an abbreviation dictionary that takes into account its recording medium and an abbreviation generation type.

The present invention is to collect original words from web documents existing on the Internet, to generate abbreviated candidates corresponding to at least one of a noun abbreviation, a syllable combination, or a mixed form for the collected original words; Calculating the probability that the abbreviation candidates are actual abbreviations by applying a frequency in which candidates exist in web documents existing on the Internet, and an abbreviation whose probability that the actual abbreviation among the abbreviation candidates is greater than or equal to an abbreviation threshold Listing the candidates in the abbreviation dictionary.

According to the present invention, there is provided a method for automatically constructing a dictionary of Korean abbreviations, providing high efficiency and high accuracy, and generating only possible abbreviation candidates in consideration of the linguistic characteristics of the Korean language, and thus used to construct an abbreviation dictionary. It is possible to effectively reduce the time required, and construct a dictionary with high accuracy and high recall using a probabilistic model using the features of the abbreviation.

Description

Method for automatic construction of acronym dictionary based on acronym type, Recording medium etc and Apparatus for automatic construction of acronym dictionary based on acronym type}

1 is a block diagram of an apparatus for automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to the present invention.

2 is a flowchart of a method of automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to an embodiment of the present invention.

3A and 3B are graphs showing the number of preferred syllables by generation type as a reason for separating the classification model of FIG. 2 by type.

4 is a flowchart of a method of automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to another embodiment of the present invention.

5 is a flowchart of a method of automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to another embodiment of the present invention.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to natural language processing, and more particularly, to a method for automatically building an abbreviation dictionary that takes into account an abbreviation generation type, and an apparatus for automatically building an abbreviation dictionary that takes into account a recording medium and an abbreviation generation type.

The abbreviation is a shortened word of syllables or morphemes. For example, abbreviation of 'domestic product' to 'domestic' or 'KEPCO' to use the abbreviation 'KEPCO' is an example of the use of the abbreviation. Acronyms are a phenomenon used for the efficiency of languages, and are especially frequent in the present day when the Internet is actively used.

Identifying the abbreviation and finding the original language of the abbreviation is required for information retrieval to search for desired information on the web or to extract information for extracting only desired information from a specific document. For example, in the case of information retrieval, when an abbreviation appears in the query of the user who wants to search, it automatically adds the abbreviation of the abbreviation to the query to increase the possibility of finding the desired document and also uses only the original word in the query. When it is done, you can improve the performance by searching for documents with abbreviations. In addition, by building a dictionary of abbreviations, it is helpful to study linguistics and linguistics, and it is possible to easily understand the original language when the abbreviation is used in the document.

However, in the case of abbreviations, the number of abbreviations already existing is very large, and constantly new ones are generated continuously.

Therefore, the conventional abbreviation dictionary is difficult to build a dictionary manually, there is a problem that can not reflect the characteristics of the Korean abbreviation, and can not automatically build an abbreviation dictionary.

Accordingly, the first technical task of the present invention is to provide a method for automatically constructing a dictionary of Korean abbreviations, provide high efficiency and high accuracy, and generate only possible abbreviation candidates in consideration of the linguistic characteristics of Korean. An automatic method for building abbreviation dictionaries that can efficiently reduce the time used to construct the algorithm, and to consider the type of abbreviation generation that can build dictionaries with high accuracy and high recall using a probabilistic model using the characteristics of the abbreviations. To provide.

A second technical object of the present invention is to provide a computer readable recording medium having recorded thereon a program for executing the automatic abbreviation dictionary automatic construction method considering the above abbreviation generation type.

The third technical problem to be achieved by the present invention is to provide a device for automatically building an abbreviation dictionary that takes into account an abbreviation generation type to which the method for automatically building an abbreviation dictionary considering the above abbreviation generation type is applied.

In order to achieve the first technical problem, the present invention comprises the steps of collecting original words from web documents existing on the Internet, corresponding to at least one form of noun abbreviation, syllable combination or mixed form for the collected original words. Generating abbreviation candidates, frequency of occurrence of the abbreviation candidates in web documents present on the Internet in the classification model that considers the qualities for distinguishing abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations Calculating a probability that the abbreviation candidates are actual abbreviations and abbreviating the abbreviation candidates having a probability that is greater than or equal to the actual abbreviation than an abbreviation threshold for selecting an abbreviation candidate having a high probability that the abbreviation candidate is not a real abbreviation. Dictionary of abbreviations that takes into account the type of abbreviation that includes listing in a dictionary It provides the same construction method.

In addition, in order to achieve the first technical problem, the present invention comprises the steps of collecting the original language from the web documents existing on the Internet, at least one form of noun abbreviation, syllable combination or mixed form for the collected original words Generating corresponding abbreviation candidates, wherein the abbreviation candidates are present in web documents present on the Internet in the classification model that considers qualities for distinguishing abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations; Calculating a probability that the abbreviation candidates are actual abbreviations and non-abbreviations by applying a frequency of the abbreviations, and the probability that the abbreviation candidates are actual abbreviations is greater than an abbreviation threshold for selecting an abbreviation candidate having a high probability of not being an abbreviation. Selecting an abbreviation candidate that is greater than or equal to, the actual drug among the selected abbreviation candidates; Further screening for abbreviation candidates that are unlikely to be actual abbreviations that are unlikely to be abbreviations and removing abbreviated candidates from the selected abbreviation candidates to increase accuracy; A method for automatically building an abbreviation dictionary that takes into account the type of abbreviation generation comprising the step of listing the listed abbreviation candidates in the abbreviation dictionary.

In addition, in order to achieve the first technical problem, the present invention comprises the steps of collecting the original language from the web documents existing on the Internet, at least one form of noun abbreviation, syllable combination or mixed form for the collected original words Generating corresponding abbreviation candidates, wherein the abbreviation candidates are present in web documents present on the Internet in the classification model that considers qualities for distinguishing abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations; Calculating a probability that the abbreviation candidates are actual abbreviations and non-abbreviations by applying a frequency of the abbreviations, and the probability that the abbreviation candidates are actual abbreviations is greater than an abbreviation threshold for selecting an abbreviation candidate having a high probability of not being an abbreviation. Selecting candidates with a greater or equal abbreviation, the probability that the actual abbreviation is the abbreviation threshold Among the smaller abbreviation candidates, the difference between the probability of actual abbreviation and the probability of non-abbreviation may be abbreviated candidates greater than or equal to the recall factor threshold to increase reproducibility by restoring an abbreviation candidate that may become an actual abbreviation. A method for automatically constructing an abbreviation dictionary considering a type of an abbreviation including adding to and listing the selected abbreviation candidates in an abbreviation dictionary.

In order to achieve the above second technical problem, the present invention provides a computer-readable recording medium having recorded thereon a program for executing the automatic abbreviation dictionary construction method in consideration of the above-mentioned abbreviation generation type.

In order to achieve the third technical problem, the present invention provides a language collection unit that collects original languages from web documents existing on the Internet, and includes at least one of a noun abbreviation, a syllable combination, or a mixed form for the collected original languages. An abbreviation candidate generation unit for generating corresponding abbreviation candidates, and a web document in which the abbreviation candidates exist on the Internet in the classification model considering a qualification for distinguishing an abbreviation as a classification model for calculating a probability that the abbreviation candidates are actual abbreviations. Classification model that calculates the probability that the abbreviation candidates are actual abbreviations by applying a frequency present in the field, and the probability that the actual abbreviation is higher than an abbreviation threshold for selecting an abbreviation candidate having a high probability that the abbreviation candidates are not the actual abbreviation. An abbreviation that includes a dictionary generator that lists abbreviated candidates greater than or equal to the abbreviation dictionary. Provides an automatic abbreviation dictionary construction device that takes into account the type of generation.

Hereinafter, with reference to the drawings will be described a preferred embodiment of the present invention. However, embodiments of the present invention illustrated below may be modified in various other forms, and the scope of the present invention is not limited to the embodiments described below.

1 is a block diagram of an apparatus for automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to the present invention.

The automatic abbreviation dictionary construction apparatus according to the present invention considers the characteristics of the abbreviation generation for a given original language as shown in FIG. 1, generates candidates for all possible abbreviations, and classifies the abbreviations actually used by using a probability model. Can build. In the case of the original language, it can be easily obtained on a website. In particular, in the case of proper nouns, it is not difficult to collect a list of original languages through a website related to a specific domain. For example, a list of company names can be easily obtained from the Korea Chamber of Commerce and Industry or the stock exchange site. Therefore, a sufficient amount of dictionaries can be constructed according to the purpose.

The original language collecting unit 110 collects original languages from web documents existing on the Internet. In this case, the network interface unit 120 may be used to connect to the Internet. For example, the network interface unit 120 may be an interface device such as an Ethernet support card or a wireless LAN card.

The abbreviation candidate generation unit 130 generates abbreviation candidates corresponding to at least one of a noun abbreviation, a syllable combination, or a mixed form with respect to the original words collected by the original language collection unit 110.

The classification model unit 140 calculates the probability that the abbreviation candidates are actual abbreviations by applying a frequency in which the abbreviation candidates generated by the abbreviation candidate generator 130 are present in web documents existing on the Internet to a predetermined classification model. . For example, the given classification model may be a simple Bayesian classification model.

The dictionary generator 150 registers the abbreviation candidates having the probability that the actual abbreviation calculated by the classification model unit 140 is greater than or equal to the abbreviation threshold among the abbreviation candidates. Listing in advance involves building a database. For example, the database may include a hard disk drive, an optical recording medium, and a recording device thereof. In this case, the abbreviation threshold value is a threshold value for selecting candidates having a high probability of not being an actual abbreviation among the abbreviation candidates, and may be determined through an iterative experiment.

For example, the automatic abbreviation dictionary construction device that takes into account the type of abbreviation generation according to the present invention is a storage means capable of constructing a large-capacity database, an I / O device connected to the storage means, performs a main operation and classifies models and stemming. It can be implemented using a computer system including a microprocessor for performing operations for the analyzer, a memory for the microprocessor, and an interface card connected to the Internet and assigned an IP address.

2 is a flowchart of a method of automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to an embodiment of the present invention.

First, original languages are collected from web documents existing on the Internet (step 210).

Next, abbreviation candidates corresponding to at least one of noun abbreviation, syllable combination, or mixed form are generated for the collected original words (step 220). Preferably, this process (step 220) is a process of analyzing the morphemes of the collected original words to divide the original words into constituent nouns and at least one form of noun abbreviation, syllable combination or mixed form for the original words divided into constituent nouns. It may include the process of generating the abbreviation candidate corresponding to the.

Once the abbreviation candidates are generated, the probability that the abbreviation candidates are actual abbreviations is calculated by applying a frequency in which the abbreviation candidates exist in web documents existing on the Internet to a predetermined classification model (step 230). Preferably, this process (230) is a process of constructing a classification model by considering characteristics specific to noun abbreviations, syllable combinations, and mixed forms, and applying abbreviated candidates to corresponding classification models among the classification models. It may include.

Finally, abbreviation candidates whose probabilities are actual abbreviations above the abbreviation threshold are listed in the abbreviation dictionary (step 240).

For example, each of the above steps is described below.

First assume that the original language has been collected, and as a first step create candidates with possible abbreviations for the given original languages. In the case of Korean abbreviations, a specific noun is generated by omitting a specific noun, such as using the Ministry of Education and Human Resources as a Ministry of Education, or combining certain syllables of a particular noun, such as using the Ministry of Information and Communication as an Orthodox Department. To generate abbreviations. In the present invention, as shown in Table 1, a method of generating abbreviations is generated and candidates for possible abbreviations are generated according to the method.

In order to generate an abbreviation according to the above classification, the present invention applies one of the following rules to each noun constituting the original language.

Rule 1. Omit nouns

Rule 2 Choose a syllable's first syllable or end syllable to participate in the abbreviation

Rule 3. Participating in the entire noun

Also, depending on the rules applied, the type of abbreviation generated can be determined as follows.

In other words, the noun abbreviation is an abbreviation generated by applying rule 1 or rule 3, for example, when "education / human / resources / part" is referred to as the "education part", and the syllable combination type uses the rule 1 or rule 2. As an abbreviation generated by this, for example, "country / information / source" is set to "National Intelligence Service". In addition, the mixed type is an abbreviation generated by applying both Rule 2 and Rule 3, for example, when "Daewoo Motor Sales" is "Daewoo Keyboard".

To generate abbreviated candidates by applying the above method, it is necessary to divide the original language into constituent nouns. To this end, the morphological analyzers developed in the present invention may be used.

As a second step, only original and abbreviation pairs can be selected by using a probability model for the original and abbreviation candidate pairs generated in the first step. Probability models that can be used include simple Bayesian classification models that are commonly used for classification. The simple Bayesian classification model uses Equation 1 to calculate the probability that the abbreviation candidate is the actual abbreviation and the probability that the abbreviation is not the abbreviation given the original language and the abbreviation candidate.

와

는 자질 집합

로 바꾸어 표현 가능 하며 이는 다시 자질집합을 구성하는 자질로 나누어

으로 나타낼 수 있다. 이를 다시 수학식 2를 사용하여 최종적으로 약어 후보가 실제 약어 인지 아닌지를 가려 낼 수 있다.

Wow

Set of qualities

Can be expressed in terms of the features that make up the feature set.

It can be represented as Equation 2 can be used again to finally determine whether the abbreviation candidate is a real abbreviation.

In Equation 1 and Equation 2, the original and abbreviated candidates are expressed as a set of qualities to determine whether the candidate word is an abbreviation, a non-abbreviation, or an abbreviation. In other words, the abbreviation is expressed as a specific state, and the probability of being an abbreviation and not of an abbreviation is calculated in this state. For example, a given source and abbreviation candidate consists of two qualities: the length of the abbreviation, the number of web documents in which the abbreviation appears, and for a particular source and abbreviation candidate (abbreviation length = 3, the number of web documents appearing = 10). In terms of), three probabilities are computed in a given document for the case where a word in this state is an abbreviation. In other words, if a given document (learning set) has 10 words of length 3 and the number of web documents that appeared 10, of which 7 are actual abbreviations, the probability of a word having such a status is an abbreviation of 0.7.

The following qualities can be used to represent a given source and abbreviation candidate. That is, the difference between the number of nouns omitted and abbreviated nouns when generating abbreviations, the number of nouns omitted when generating abbreviations, the number of syllables in the abbreviation, the number of syllables in the original and abbreviation, the number of selected ending syllables, and the mixed form. The number of syllables combined, the abbreviations in the web document, and the frequency of the original language.

For example, the difference between the number of syllables in the original and abbreviations is as follows. In the case of the abbreviation 'KEPCO', one unit noun 'corporation' was omitted from the original 'KEPCO', and two syllables 'Gong' and 'Sa' may be omitted. Therefore, in consideration of both syllables and unit nouns as the unit of abbreviation, it is possible to construct the qualities and determine whether the actual abbreviation candidate is an abbreviation considering the length of abbreviation and the length of the abbreviation.

For example, the number of selected end syllables is as follows. When selecting syllables from syllable combination type and mixed type, it is common to select the first syllable of unit noun like KEPCO, which is abbreviation of KEPCO, and not to select the last syllable. Therefore, consider the number of ending syllables included in the abbreviation.

For example, the number of syllables combined in a mixed form is as follows. Consider how many specific syllables are used when extracting and combining specific syllables of unit nouns in a blend.

In the case of abbreviations, it is possible to find a tendency to appear in the web document at the same time as the original language, while on the other hand, it is difficult to find an affirmative document unless the abbreviation candidate is used as the abbreviation of the corresponding original language. Therefore, it is possible to find a web document in which the abbreviation and the original language appear at the same time and determine whether the abbreviation is an actual abbreviation through the air frequency.

Web documents are collected using information retrieval, but two types of documents are collected and used for each query: the abbreviation candidate and the original language as a query, and the abbreviation candidate only as a query.

3A and 3B are graphs showing the number of preferred syllables by generation type as a reason for separating the classification model of FIG. 2 by type.

In the case of the Korean abbreviation, characteristics such as length are different for each generation type. For example, as shown in FIGS. 3A and 3B, it can be seen that the number of preferred syllables is different for each generation type. In addition, specific qualities can be applied only to specific types (for example, #OfRepChar). Therefore, the learning set can be separated by generation type, and the model can be expected to improve performance by changing the qualities to be considered. In the following, the performance of each model is evaluated by applying the two types of models that do not consider the type and when the model is separated in consideration of the generation type.

For the experiment, more than 314 original languages were used, consisting of company name, national organization name and university name. Among 314 original languages, 835 original and abbreviation pairs for 307 original languages were used as positive examples, and 50139 inappropriate original and abbreviation pairs that could be generated in 314 original languages were constructed as negative examples.

In the following description, AbbNum is the number of nouns omitted when generating an abbreviation, DiffNoun is the difference between the number of nouns and nouns omitted when generating an abbreviation, AcroLen is the number of syllables in an abbreviation, and DiffLen is the difference between the number of syllables in the original and abbreviations. , #ofLastChar is the number of selected end syllables, #ofRepChar is the combined number of syllables in a mixed type, CoOccurFreq is the frequency of air in the collected documents, and DefFreq is the abbreviation of the original frequency in the searched document. Means.

5 Fold Cross Validation was used for evaluation and learning, and performance was measured by the average of each result. The accuracy rate and recall rate for the evaluation are calculated as in Equation 3 below.

Table 2 shows the performance of adding each feature when the model is not divided by generation type.

endowment Accuracy Recall F maesure Baseline 47.91% 68.60% 56.41% AbbNum 52.71% 67.20% 59.08% DiffLen 50.15% 67.70% 57.62% #OfLastChar 58.78% 69.60% 63.74% all 58.68% 73.00% 65.06%

Table 3 shows the performance when considering the abbreviation generation type according to the present invention.

type endowment Accuracy Recall F-Measure Noun DiffNoun 72.58% 69.33% 70.92% Syllable combination #OfLastChar + DefFreq 67.47% 68.29% 67.88% Mixed DiffLen + # OfLastChar + # OfRepChar 51.67% 79.45% 62.62% Integrated performance 64.52% 72.00% 68.05%

As a baseline, we use the performance of a classifier that uses only the frequency of the source and abbreviations in the web document as a feature without considering other characteristics. Table 1 shows the performance of adding useful features to the baseline to improve the performance of the baseline. It can be seen that the performance is improved by about 8.4% compared to the baseline. Table 2 shows the performance of different features and learning sets according to the type of abbreviation, the useful features, and the integrated performance of the entire learning set. In the case of nominal abbreviations, there is no meaningful level of learning set, so performance is measured only when the model is not divided by type. It can be seen that the performance is improved by about 3% compared to when the generation type is not considered.

Many of the cases in which words that are not actual abbreviations are classified as abbreviations are those when misspelled words appear on the web. Considering that many of these words are missing from intermediate Hangul characters such as (Daewoo Motor Sales, Daewoo Auto Sales), the candidates are eliminated to obtain 70.51% accuracy, 71% recall, and 70.75% F-measure. Can be.

In the case of an abbreviation dictionary, accuracy may be important rather than recall, and vice versa, depending on the purpose. Therefore, the need to adjust this is required. In the present invention, after normalizing the probability result value of the classifier, if the difference between the probability of the abbreviation and the probability that is not the abbreviation among the candidates judged as the abbreviation is not more than a certain level, the accuracy rate is increased and the reproducibility is lowered by not determining the abbreviation. Can be. Below is the formula for judging the abbreviation in this way.

In order to reduce the accuracy and increase the reproducibility, on the contrary, if the difference in the probability value exceeds a certain value for a candidate that is not determined as an abbreviation, it is determined as an abbreviation.

Table 4 shows the accuracy, recall and F-measure values when adjusted in this manner.

Accuracy Recall F-Measure Increase in accuracy 82.30% 54.40% 65.50% Higher recall 53.38% 83.00% 64.97%

4 is a flowchart of a method of automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to another embodiment of the present invention.

First, original languages are collected from web documents existing on the Internet (step 410).

Next, abbreviation candidates corresponding to at least one of noun abbreviation, syllable combination, or mixed form are generated for the collected original words (step 420).

Once the abbreviation candidates are generated, the frequency at which the abbreviation candidates are present in the web documents existing on the Internet is applied to a predetermined classification model to calculate the probability that the abbreviation candidates are actual abbreviations and not the abbreviations (step 430).

Next, among the abbreviation candidates, the abbreviation candidates whose probability of being the actual abbreviation is greater than or equal to the abbreviation threshold is selected (step 440).

Once the abbreviation candidates are selected, abbreviated candidates whose difference between the probability of being the actual abbreviation and the non-abbreviation of the selected abbreviation candidates is below the accuracy factor threshold are removed from the selected abbreviation candidates (steps 450, 460).

In this case, the accuracy rate threshold value is a threshold value for increasing the accuracy rate by additionally selecting candidates that are not likely to be actual abbreviations among the abbreviation candidates and may be determined through an iterative experiment.

Finally, the selected abbreviation candidates are listed in the abbreviation dictionary (step 470).

5 is a flowchart of a method of automatically building an abbreviation dictionary in consideration of an abbreviation generation type according to another embodiment of the present invention.

First, original languages are collected from web documents existing on the Internet (step 510).

Next, abbreviation candidates corresponding to at least one of noun abbreviation, syllable combination, or mixed form are generated for the collected original words (step 520).

Once the abbreviation candidates are generated, the frequency with which the abbreviation candidates are present in the web documents existing on the Internet is applied to a predetermined classification model to calculate the probability that the abbreviation candidates are actual abbreviations and not the abbreviations (step 530).

Next, among the abbreviation candidates, the abbreviation candidates whose probability of being the actual abbreviation is greater than or equal to the abbreviation threshold is selected (step 540).

When the abbreviation candidates are selected, the abbreviation candidates selected as non-abbreviation candidates above, that is, the difference between the probability that the actual abbreviation is less than the abbreviation threshold and the probability that the actual abbreviation is not the abbreviation is greater than the recall rate (Steps 550, 560).

In this case, the reproducibility threshold may be determined through repetitive experiments as a threshold for increasing the reproducibility by restoring candidates that may be abbreviations even though it is determined that the abbreviation is not an abbreviation.

Finally, the selected abbreviation candidates are listed in the abbreviation dictionary (step 570).

Preferably, a method for automatically building an abbreviation dictionary in consideration of the abbreviation generation type of the present invention may be provided by recording a program for executing on a computer on a computer-readable recording medium.

The invention can be implemented via software. When implemented in software, the constituent means of the present invention are code segments that perform the necessary work. The program or code segments may be stored on a processor readable medium or transmitted by a computer data signal coupled with a carrier on a transmission medium or network.

The computer-readable recording medium includes all kinds of recording devices in which data is stored which can be read by a computer system. Examples of computer-readable recording devices include ROM, RAM, CD-ROM, DVD ± ROM, DVD-RAM, magnetic tape, floppy disks, hard disks, optical data storage devices, and the like. The computer readable recording medium can also be distributed over network coupled computer devices so that the computer readable code is stored and executed in a distributed fashion.

Although the present invention has been described with reference to one embodiment shown in the drawings, this is merely exemplary and will be understood by those of ordinary skill in the art that various modifications and variations can be made therefrom. However, such modifications should be considered to be within the technical protection scope of the present invention. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

As described above, the present invention provides a method for automatically constructing a dictionary of Korean abbreviations, provides high efficiency and high accuracy, and generates only possible abbreviation candidates in view of the linguistic characteristics of Korean. It can effectively reduce the time used for constructing and construct a dictionary with high accuracy and high recall using a probabilistic model using the features of the abbreviations.

Claims (13)

Collecting original languages from web documents existing on the Internet; Generating abbreviated candidates corresponding to at least one of a noun abbreviation, a syllable combination, or a mixed form for the collected original words; The abbreviation candidates are applied by applying the frequency of the abbreviation candidates in web documents existing on the Internet to the classification model that considers the qualities for distinguishing the abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations. Calculating probabilities that are actual abbreviations; And An abbreviation dictionary that takes into account an abbreviation generation type comprising listing an abbreviation candidate having a probability that is greater than or equal to the actual abbreviation than an abbreviation threshold for screening an abbreviation candidate having a high probability of not being an actual abbreviation. Auto build method. The method of claim 1, Generating the abbreviation candidates Analyzing the morphemes of the collected original languages and dividing the original languages into constituent nouns; And Generating abbreviation candidates corresponding to at least one of noun abbreviation, syllable combination, or mixed form for the original words divided into the constituent nouns. . The method of claim 1, Computing the probability that the abbreviation candidates are actual abbreviations Constructing a classification model for a noun abbreviation, a syllable combination, and a mixture, respectively; And And applying the abbreviation candidates to a corresponding classification model among the constructed classification models. The method of claim 3, wherein Comprising the classification model The classification model for the noun abbreviation is a feature for distinguishing the abbreviation, the model for considering the abbreviation generation type, characterized in that the model to consider the number of nouns omitted when generating the abbreviation. The method of claim 3, wherein Comprising the classification model The classification model for the sound quality combination type is an abbreviation for considering the type of abbreviation, which is a model that considers the frequency of appearance of original words in a searched document by querying the number of end syllables and the candidate candidates as a query to distinguish the abbreviations. Pre-automatic build method. The method of claim 3, wherein Comprising the classification model The classification model for the hybrid type is a feature for distinguishing the abbreviation, considering the difference between the number of syllables of the original and the abbreviation, the number of selected end syllables, and the number of combined syllables in the mixed type. An acronym dictionary auto-building method The method of claim 1, Computing the probability that the abbreviation candidates are actual abbreviations Method for automatically building an abbreviation dictionary considering the type of abbreviation, characterized in that the classification model is a simple Bayesian classification model. Collecting original languages from web documents existing on the Internet; Generating abbreviated candidates corresponding to at least one of a noun abbreviation, a syllable combination, or a mixed form for the collected original words; The abbreviation candidates are applied by applying the frequency of the abbreviation candidates in web documents existing on the Internet to the classification model that considers the qualities for distinguishing the abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations. Calculating probabilities that are actual abbreviations and probabilities that are not abbreviations; Selecting an abbreviation candidate having a probability greater than or equal to the actual abbreviation than an abbreviation threshold for selecting an abbreviation candidate having a high probability of not being an actual abbreviation among the abbreviation candidates; Among the selected abbreviation candidates, the candidates having a smaller probability than the accuracy factor threshold for increasing the accuracy by further selecting an abbreviation candidate having a high probability that the difference between the probability of being the actual abbreviation and the probability that is not the abbreviation are not the actual abbreviation are selected. Removing from abbreviation candidates; And And automatically listing the selected abbreviation candidates in an abbreviation dictionary. The method of claim 8, Computing the probability that the abbreviation candidates are actual abbreviations Constructing a classification model for a noun abbreviation, a syllable combination, and a mixture, respectively; And And applying the abbreviation candidates to a corresponding classification model among the constructed classification models. Collecting original languages from web documents existing on the Internet; Generating abbreviated candidates corresponding to at least one of a noun abbreviation, a syllable combination, or a mixed form for the collected original words; The abbreviation candidates are applied by applying the frequency of the abbreviation candidates in web documents existing on the Internet to the classification model that considers the qualities for distinguishing the abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations. Calculating probabilities that are actual abbreviations and probabilities that are not abbreviations; Selecting an abbreviation candidate having a probability greater than or equal to the actual abbreviation than an abbreviation threshold for selecting an abbreviation candidate having a high probability of not being an actual abbreviation among the abbreviation candidates; The difference between the probability of the actual abbreviation and the probability of not being the abbreviation among the abbreviation candidates whose probability of being the actual abbreviation is smaller than the abbreviation threshold is greater than the recall factor threshold for increasing the reproducibility by restoring the abbreviation candidate that can be the actual abbreviation. Or adding an abbreviation candidate to the selected abbreviation candidate; And automatically listing the selected abbreviation candidates in an abbreviation dictionary. The method of claim 10, Computing the probability that the abbreviation candidates are actual abbreviations Constructing a classification model for a noun abbreviation, a syllable combination, and a mixture, respectively; And And applying the abbreviation candidates to a corresponding classification model among the constructed classification models. A computer-readable recording medium having recorded thereon a program for executing the method of claim 1 on a computer. A language collection unit for collecting the languages from web documents existing on the Internet; An abbreviation candidate generator for generating abbreviation candidates corresponding to at least one of a noun abbreviation, a syllable combination, or a mixed form with respect to the collected original words; The abbreviation candidates are applied by applying the frequency of the abbreviation candidates in web documents existing on the Internet to the classification model that considers the qualities for distinguishing the abbreviations as a classification model for calculating the probability that the abbreviation candidates are actual abbreviations. A classification model unit for calculating a probability of being an actual abbreviation; And An abbreviation considering an abbreviation generation type including a dictionary generation unit for listing an abbreviation candidate having a probability that is greater than or equal to the actual abbreviation than an abbreviation threshold for selecting an abbreviation candidate having a high probability of not being an actual abbreviation among the abbreviation candidates. Pre-automatic build device.

Images