KR20100138194A - System and method for recommendding japanese language automatically using tranformatiom of romaji - Google Patents

Abstract

PURPOSE: A system and a method for automatically recommending Japanese using conversion of roman letters are provided to convert a pronunciation of an inputted Japanese word into Roman letters and search similarity of the word based on the converted Roman letters. CONSTITUTION: A Roman letter converting unit(103) converts a pronunciation of a word expressed by hiragana or katakana into Roman letters. A similar word search unit(104) searches a similar word about the word based on the converted Roman letters. The similar word search unit searches a similar word for the word in consideration of a similarity score of the word which is converted into Roman letters. A similar word recommending unit(105) converts the searched similar word into one form of hiragana, katakana, and Chinese characters.

Description

Japanese automatic recommendation system and method using roman conversion {SYSTEM AND METHOD FOR RECOMMENDDING JAPANESE LANGUAGE AUTOMATICALLY USING TRANFORMATIOM OF ROMAJI}

The present invention relates to a system and method for recommending a similar word for the input Japanese, and more particularly, to a system and a method for recommending the similar word by converting the pronunciation of the input Japanese to Roman characters.

The user enters a word into the search window of the search engine to perform the search to obtain the desired information. At this time, when a user incorrectly inputs a word and a typo occurs, the quality of the searched document is reduced due to the typo, or the number of searched documents is hardly generated. In order to solve this problem, the search engine considers these words as typos and recommends the words that the user actually inputs.

In addition, even if a user searches by entering a word, only a few cases may input an optimal word for obtaining a desired result. In this case, even if the search engine provides the search results to the user, the user has to complain about the search results. In order to solve this problem, the search engine may improve the accuracy of the search by providing an associated word or a similar word for a word input by the user.

In particular, the above mentioned situations may be more problematic than in the case of Japanese search. When the Japanese inputted by the user is judged as a typo, the correct answer is provided, or when the user inputs a similar word for the Japanese input, it is difficult to guarantee the accuracy in the past. Above all, since Japanese can be expressed in the form of kanji, hiragana and katakana, it has been difficult to recommend an appropriate word for a word input by a user. Therefore, there is a need for a method for recommending an appropriate word even when Japanese is input in any form.

The present invention can provide a system and method for improving the accuracy of a similar word search for Japanese by converting a pronunciation of an input Japanese word into a roman letter and searching for a similar word for a word based on the converted roman letter.

The present invention provides a system and method for determining whether an input Japanese word is a typo, and searching for similar words in case of a typo, thereby providing a correct answer word, thereby recommending an appropriate correct word even when a user incorrectly inputs a search query. Can provide.

The present invention is a system and method that can perform fast and accurate Kanji-Hiragana conversion by dividing the token into tokens through the learning data generated by machine learning and converting the split tokens into hiragana when the entered Japanese words are Chinese characters. Can be provided.

The present invention can provide a system and method for searching and recommending similar words in a form different from a Japanese word input by a user, thereby enabling a user to perform a more accurate search.

In the Japanese automatic recommendation system according to an embodiment of the present invention, a roman conversion unit for converting a pronunciation of a word expressed in Japanese hiragana form or katakana form into a romaji and a similar word for the word based on the converted roman letter It may include a similar word search unit for searching for.

The automatic Japanese recommendation system according to an embodiment of the present invention may further include a similar word recommendation unit which converts the searched analogous words into any one of the Japanese form of hiragana, katakana or kanji.

The automatic Japanese recommendation system according to an embodiment of the present invention may further include a typo determination unit that analyzes the input word and determines whether the word is a typo.

When the input word is a typo, the Japanese automatic recommendation system according to an embodiment of the present invention selects a correct answer word for selecting the correct answer word for the word among the searched similar words in consideration of the editing distance according to the similarity score or the frequency of word appearance. It may further include wealth.

In the Japanese automatic recommendation system according to an embodiment of the present invention, when the input word is a Chinese character, the word is divided by token using token division learning data, and the divided token is used by using the Kanji-Hiragana conversion learning data. The apparatus may further include a kanji-hiragana conversion unit for converting to a corresponding hiragana.

The automatic Japanese recommendation method according to an embodiment of the present invention converts a pronunciation of a word expressed in Japanese hiragana form or katakana form into a romaji and searches for a similar word for the word based on the converted roman letter. It may include the step.

According to an embodiment of the present invention, the accuracy of the similarity search for Japanese may be improved by converting the pronunciation of the input Japanese word into a Roman letter and searching for a similar word for the word based on the converted Roman letter.

According to an embodiment of the present invention, by determining whether an input Japanese word is a typo, and searching for a similar word when a typo is provided, the correct word is recommended, even when a user incorrectly inputs a search query, by recommending an appropriate correct word to improve the accuracy of the search. Can be improved.

According to an embodiment of the present invention, when the input Japanese word is a Chinese character, the character data is divided into tokens through learning data generated through machine learning, and the divided tokens are converted into hiragana to perform fast and accurate Kanji-Hiragana conversion. can do.

According to an embodiment of the present invention, by searching for and recommending similar words in a form different from the Japanese word input by the user, the user can perform a more accurate search.

Hereinafter, with reference to the contents described in the accompanying drawings will be described in detail an embodiment according to the present invention. However, the present invention is not limited to or limited by the embodiments. Like reference numerals in the drawings denote like elements.

1 is a block diagram showing the overall configuration of a Japanese automatic recommendation system according to an embodiment of the present invention.

Referring to FIG. 1, the Japanese automatic recommendation system 100 includes a typo determining unit 101, a kanji-hiragana conversion unit 102, a roman character conversion unit 103, a similar word search unit 104, and a similar word recommendation unit 105. And a correct word selector 106.

In Japanese search, the user can enter Japanese to search for desired information. At this time, the user may input the word A 107 which is Japanese in the form of a kanji, hiragana or katakana. The Japanese automatic recommendation system 100 may recommend the more accurate Japanese word B 108 by converting the pronunciation of the word 107 input by the user into a Roman letter.

According to an embodiment of the present invention, when the user inputs a typo, the Japanese automatic recommendation system 100 may select and provide a correct answer for the typo through the typo determination unit 101 or the correct word selection unit 106. Can be. In addition, according to another embodiment of the present invention, when the user inputs a sperm other than a typo, the Japanese automatic recommendation system 100 uses the kanji-hiragana conversion unit 102 to the similar word recommendation unit 105. Can provide. Hereinafter, a description will be given focusing on a case where a user inputs a typo.

The typo determination unit 101 may analyze the word 107 input by the user to determine whether the word 107 is a typo. In this case, when the word 107 input by the user is a typo, the roman character converting unit 103 may convert the word 107 into a roman character.

For example, the typo determining unit 101 may determine whether the word 107 is a typo in consideration of whether or not the word 107 input by the user is included in the preset typo data. In detail, the typo determination unit 101 includes a word 107 input by the user in the typo data by using a pre-registered word, a typo data determined by a content DB list constructed by a search engine, and manual inspection. In this case, it can be judged as a typo.

As another example, the typo determining unit 101 may determine whether the word 107 is a typo in consideration of whether an input frequency or a document appearance frequency of the word 107 input by the user is lower than a preset reference frequency. have.

At this time, the frequency of input of the word 107 means the number of times of input of the word 107 input by the user. That is, the typo determination unit 101 may determine the word 107 having a low input frequency as a typo. In addition, the document appearance frequency may mean the number of documents derived as a search result when searching through the input word 107. That is, the typo determination unit 101 may determine the word 107 having a low document appearance frequency as a typo.

Alternatively, when the document appearance frequency of the word 107 is lower than the query frequency, the typo determination unit 101 may determine the word 107 as a typo. In addition, the typo determination unit 101 may determine the continuous word 107 as a typo while the document appearance frequency is low.

As another example, the typo determining unit 101 may determine whether the word 107 is a typo in consideration of whether the word 107 input by the user is divided into morphemes. In this case, when the input word is separated into each morpheme by a morpheme analyzer or a part-of-speech tag, the typo determination unit 101 may determine that the corresponding word 107 is not a typo. In other words, if the word is not a typo, the typos may not be easily separated into morphemes, and thus the typo determination unit 101 may determine sperm when the words are easily separated into morphemes.

If the input word 107 is a Chinese character, the Chinese-Hiragana conversion unit 102 may divide the word for each token using token split learning data. The kanji-hiragana conversion unit 102 may convert the kanji-hiragana conversion into hiragana corresponding to the divided token using the kanji-hiragana conversion training data. Since the same kanji is read differently depending on the same kanji in Japan, it is important to convert it to the correct hiragana. The kanji-hiragana conversion unit 102 will be described in detail with reference to FIG. 3.

The roman conversion unit 103 may convert a pronunciation of the word 107 expressed in a Japanese hiragana form or a katakana form into a roman character. If the word is a Chinese character, the word may be converted into hiragana through the kanji-hiragana conversion unit 102 and then converted into roman characters. For example, if the input word is 映? (Movie), which is a Chinese character, it is converted into え い 한 through the Chinese-Hiragana conversion unit 102, and the Roman conversion unit 103 converts the pronunciation of the word to Roman letters (eiga). can do. An example in which the Roman conversion unit 103 converts to Roman characters is described in detail with reference to FIG. 4.

The similar word search unit 104 may search for similar words for the word 107 based on the converted roman characters. For example, the similarity search unit 104 may search for similarity with respect to a word in consideration of a similarity score of a word converted to Roman characters. Since measuring the similarity in the hiragana / katakana or kanji state is very low in accuracy of the editing distance, the accuracy can be measured by converting the pronunciation of a word to roman. For example, it is possible to compare origon and orikon more accurately by converting it to Roman characters and comparing origon and orikon rather than directly comparing Orikon and Orikon.

In this case, the similarity score may be determined based on at least one of an input frequency according to the length of the word, an edit distance according to whether the word includes a long sound, a middle point, a tactile sound or a sound sound or a degree of comparison of the word's original state. For example, when a word is a Chinese character, the similar word search unit 104 may determine a similarity score in consideration of a comparison result of a form converted to Roman characters, a comparison result of a form converted to Hiragana, and a comparison result of the original form of a Chinese character. The similar word search will be described in detail with reference to FIG. 2.

The similar word recommendation unit 105 may convert the searched similar word into a Japanese word 108 in any one of the hiragana letter, katakana character, and kanji. The user may enter the suggested word 108 to perform a search.

For example, the similar word recommender 105 may convert the searched similar word into a word 108 having a form different from the Japanese form of the word 107 input by the user and recommend the similar word. For example, even if a user inputs a hiragana type word 107, the similar word recommender 105 may convert a similar word for the input word 107 into a Chinese character type word 108 and recommend the same.

When the word 107 input from the user is a typo, the correct word selecting unit 106 selects the correct word 108 for the word 107 among the similar words searched in consideration of the similarity score or the editing distance according to the frequency of the word input. You can choose. That is, when a plurality of similar words are recommended for a typo in the input word 107, the correct word selecting unit 106 may select and provide a similar word having a high similarity score or a high input frequency as the correct word 108. have.

FIG. 2 is a diagram illustrating a process of automatically recommending Japanese through roman conversion for input words according to an embodiment of the present invention.

When a word made in Japanese is input from the user, the typo determining unit 101 may determine whether the input word is a typo. As described above, the typo determination unit 101 determines whether the word is included in the preset typo data, whether the word input frequency or document appearance frequency is lower than the preset reference frequency, or whether the word is morphologically separated. Consideration can be made to determine whether a word is a typo.

If the input word is a typo, the correct word selecting unit 106 may select and provide a correct answer word among similar words of the input word. On the contrary, if the input word is a sperm, the correct word selection unit 106 does not operate.

As shown in FIG. 2, the input word may be one of hiragana form, katakana form, or kanji form. In this case, when the input word is in hiragana form or katakana form, the roman conversion unit 103 may convert a pronunciation of a word expressed in Japanese hiragana form or katakana form into romaji.

If the input word is in the form of a Chinese character, since it is difficult to directly convert a Chinese character to a Roman character, the Chinese character-hiragana conversion unit 102 may undergo a process of normalizing the hiragana form. Specifically, the kanji-hiragana conversion unit 102 may divide the kanji into tokens using the token split learning data and convert the kanji into hiragana corresponding to the split tokens using the kanji-hiraga conversion data. Then, the roman conversion unit 103 may convert the converted hiragana pronunciation into a roman alphabet.

Then, the similar word search unit 104 may search for similar words for the word based on the converted roman characters. In detail, the similar word search unit 104 may search for similar words with respect to a word in consideration of a similarity score of a word converted to Roman characters.

For example, the similarity score may be determined based on at least one of an input frequency according to a word length, an edit distance according to whether a word includes a long sound, a middle point, a tactile sound, or a tactile sound or a degree of comparison of a word's original state.

Word length, information -information

Long sound: ハ ロ ワ-ク (Ota), ハ ロ-ワ-ク (Ota), ハ ロ-ワ-ク (correct)

Key points: ピ-ト ロ-ズ (correct), ピ-ト ロ-ズ (correct)

Halftone: オ リ ゴ ン (Ota), オ リ コ ン (correct)

Tactile: BicCamera (Ota) BicCamera (correct)

Prototype: Flower Yori Yoshino Foinin (Ota) Flower Yori Yoko Foinin (correct)

Since the shorter the word length, the higher the frequency of input of the word, the similarity search unit 104 may increase the similarity score as the shorter word length.

Since the Japanese long sound (-) is easily inserted or deleted in comparison with other characters, the similar word search unit 104 may increase the similarity score by weighting the editing distance small when the long sound is included in the word. In addition, since the Japanese midpoint is easily inserted or deleted in comparison with other characters, the similarity search unit 104 may increase the similarity score by weighting the edit distance small when the midpoint is included in the word. . Since the Japanese tactile っ is easily omitted or misused by similar pronunciation, the similarity search unit 104 may increase the similarity score by weighting the editing distance small when the word includes the tactile sound.

In addition, the similar word search unit 104 may be reflected in the similarity score by comparing the circular state of the word, as well as the form converted to Roman characters. Comparing the primitive states can compensate for errors in the search for similar words in the Roman normalized state. For example, if the input word is う と ん (utong), the similarity search unit 104 assigns a similarity score to う ど ん (udon) having a similar circular state rather than う ろ ん (oolong), thereby providing similarity through the roman conversion. Error in judgment can be compensated for.

For example, when a word is a Chinese character, the similar word search unit 104 may determine a similarity score in consideration of a comparison result of a form converted to Roman characters, a comparison result of a form converted to Hiragana, and a comparison result of the original form of a Chinese character. Specifically, when the word is a Chinese character, the similar word search unit 104 may determine the similarity score according to Equation 1 below.

SCORE (q, t) = a * score_romaji (q_romaji, t_romaji) + b * score_hiragana

(q_hiragana, t_hiragana) + c * score_original (q_original, t_original)

Here, q means Japanese (query word) input by the user, and t means analogous word. And a, b, and c represent constants. In this case, a, b, and c may be derived through machine learning.

When a similar word is searched through this process, as shown in FIG. 2, the similar word recommender 105 may convert the searched similar word into a Japanese form of hiragana, katakana or kanji. For example, when the input word is in hiragana form, the similar word recommendation unit 105 may convert the searched similar word into any one of hiragana form, katakana form, or kanji form and recommend it. That is, the similar word recommendation unit 105 may convert the searched similar word into a form different from the Japanese form of the input word and recommend the same.

For example, the similar word recommendation unit 105 may not recommend the similar word when the difference between the similarity between the state converted to the roman alphabet and the similarity between the state not converted to the Roman alphabet exceeds a preset criterion. As another example, the similar word recommendation unit 105 may not recommend the similar word when the input word is used more than the recommended similar word.

In addition, when the input word is a typo, the correct word selecting unit 106 may select the correct answer word for the word from among similar words searched in consideration of the similarity score or the editing distance according to the frequency of input of the word. In detail, the correct word selecting unit 106 may select the correct word for the word from the similar word having the highest similarity score or the high frequency of input of the word and the low editing distance.

3 is a diagram illustrating a process of converting a Chinese character into hiragana according to an embodiment of the present invention.

Kanji-Hiragana conversion unit according to an embodiment of the present invention may convert the input Chinese characters to hiragana. Then, the roman conversion unit may convert hiragana to roman characters.

For example, the kanji-hiragana conversion unit divides the kanji 304 by token according to the token division 305 using the token division learning data 302 and the kanji-hiragana using the kanji-hiragana conversion learning data 303. The transform 306 may convert the split token 305 into a hiragana 307 corresponding to the split token 305.

In the case of 僕 と 彼女 の 生 き る 道, the tokens are divided into 僕, と, 彼女, の, 生 き, る, and 道 using the token literacy learning data (302), and the hiragana state sequence has the maximum probability value in each toeogram. Select. As a result, it can be performed as follows.僕-ぼ く と 彼女-か の じ ょ の 生 き る-い き る 道-み ち Finally it can be converted to ぼ く と か の じ ょ の い き る み ち.

At this time, the learning data is made from a Japanese document 301, such as a document published in Japanese news or Japanese blog, Hiragana learning document corresponding to the kanji 304, and based on the learning document to the input form through the machine learning method By selecting and combining the hiragana according.

For example, the token split learning data 302 may be determined through a hidden markov model-based spacing learning using a corpus divided by morpheme tokens of Chinese characters. In this case, the token split learning data 302 may be determined through the spaced learning based on the syllable trigam HMM.

In one example, the Hanja-Hiragana conversion training data 303 is a unigram dictionary 303-1 and a bigram dictionary determined through learning based on a corpus separated by morpheme tokens of the Chinese character 304. 303-2. At this time, the unigram dictionary 303-1 may be constructed as a frequency (token-hiragana) between the token and the hiragana. The viagram dictionary 303-2 may be constructed with a frequency between tokens (token 1-token 2). That is, the kanji-hiraga conversion unit may convert the kanji 304 into the hiragana 307 using the token split learning data 302 and the kanji-hiragana conversion learning data 303 determined from the document 301 through the learning process. have.

For example, the Chinese-Hiragana conversion unit may select a token having a maximum probability by searching the bigram dictionary 303-2 for each token divided from the Chinese character 304 through the token split learning data 301. . The kanji-hiragana conversion unit 102 may convert the finally selected tokens into the hiragana 307 corresponding to the unigram dictionary 303-1. If the amount of information of the bigram dictionary 303-2 is insufficient, the Hanja-Hiragana converter may select the token having the maximum probability using the unigram dictionary 303-1.

4 is a diagram illustrating an example of converting to Roman characters according to an embodiment of the present invention.

According to FIG. 4, the example which converts to "a" line and "a" line to Roman characters is shown. The roman conversion unit may convert a pronunciation of a word expressed in Japanese hiragana form or katakana form into romaji. In this case, when the input word is a Chinese character, the Chinese character-Hiragana conversion unit may convert the Chinese character to Hiragana.

As can be seen in FIG. 4, for the A action, the Roman conversion part may convert A to the Roman letter “a”. The roman conversion unit may convert i to roman "i". Similarly, the Roman conversion unit can convert う to "u", え to "e", and お to "o", respectively. Through this process, the Japanese automatic recommendation system can search for similar words of the words inputted more precisely by converting Hiragana or Katakana to Roman characters.

As described above, when searching for similar words using hiragana and katakana as it is, the resolution of the editing distance is low, so it is difficult to distinguish between the original and the original in a machine such as a non-human server. In this case, the accuracy of the similarity recommendation can be improved by calculating a more precise similarity score by comparing the orignon and the oricon to the roman letters origon and orikon.

5 is a flowchart illustrating the entire process of the Japanese automatic recommendation method according to an embodiment of the present invention.

Referring to FIG. 5, the Japanese automatic recommendation system may determine whether a word input from the user is a typo (S501). In this case, when the input word is a typo, the Japanese automatic recommendation system may select and provide a correct answer word among similar words for the word (S507).

The Japanese automatic recommendation system may automatically recommend similar words for the input words even if the input words are sperm as well as typos. The Japanese automatic recommendation system may determine whether the input word is a kanji (S502). If the word is a kanji, the Japanese automatic recommendation system may convert the kanji to hiragana (S503). Then, step S504 is performed. If the entered word is not a Chinese character, no separate conversion process is performed.

In detail, the Japanese automatic recommendation system may divide the word into tokens by using token division learning data, and convert the word into hiragana corresponding to the divided tokens using the kanji-hiragana conversion learning data.

In this case, the token split learning data may be determined through a hidden markov model based on Hidden Markov Model (HMM) using a corpus divided by morpheme tokens of Chinese characters. In addition, the Chinese-Hiragana conversion training data may include a bigram dictionary and a unigram dictionary determined through learning based on a corpus separated by morpheme tokens of the Chinese character. Here, the bigram dictionary may be constructed with a frequency between tokens, and the unigram dictionary may be constructed with a frequency between tokens and hiragana.

Then, the Japanese automatic recommendation system may search the bigram dictionary for the divided tokens, select a token representing the maximum probability, and convert the hiragana corresponding to the unigram dictionary for the selected token.

The Japanese automatic recommendation system may convert the pronunciation of a word expressed in Japanese hiragana form or katakana form to Roman characters (S504). Then, the Japanese automatic recommendation system may search for similar words for the word based on the converted roman characters (S505).

For example, the Japanese automatic recommendation system may search for a similar word for a word in consideration of a similarity score of a word converted to Roman characters. In this case, the similarity score may be determined based on at least one of an input frequency according to the length of the word, an edit distance according to whether the word includes a long sound, a midpoint, a tactile sound, or a tactile sound or a degree of comparison of the word's original state.

In addition, the Japanese automatic recommendation system may convert the searched analogous words into any Japanese form of hiragana, katakana, or kanji (S506). In this case, the similar word recommendation unit may recommend the similar word searched by converting the searched similar word into a form different from the Japanese form of the input word.

For example, when the difference between the similarity of the state converted to Roman characters and the similarity of the state not converted to Roman characters exceeds a preset criterion, the Japanese automatic recommendation system may not recommend similar words. As another example, when the input word is used more than the recommended similar word, the Japanese automatic recommendation system may not recommend the similar word.

When the word input in step S501 is determined as a typo, the Japanese automatic recommendation system may select the correct answer word for the word from among the similar words searched in consideration of the similarity score or the edit distance according to the frequency of word appearance (S507).

Portions not specifically described in FIG. 5 may refer to descriptions of FIGS. 1 to 4.

In addition, the Japanese automatic recommendation method according to an embodiment of the present invention includes a computer readable medium including program instructions for performing operations implemented by various computers. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. The media may be program instructions that are specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above-described embodiments, which can be variously modified and modified by those skilled in the art to which the present invention pertains. Modifications are possible. Accordingly, the spirit of the present invention should be understood only by the claims set forth below, and all equivalent or equivalent modifications thereof will belong to the scope of the present invention.

1 is a block diagram showing the overall configuration of a Japanese automatic recommendation system according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a process of automatically recommending Japanese through roman conversion for input words according to an embodiment of the present invention.

3 is a diagram illustrating a process of converting a Chinese character into hiragana according to an embodiment of the present invention.

4 is a diagram illustrating an example of converting to Roman characters according to an embodiment of the present invention.

5 is a flowchart illustrating the entire process of the Japanese automatic recommendation method according to an embodiment of the present invention.

<Explanation of symbols for the main parts of the drawings>

100: Japanese automatic recommendation system

101: typo judgment

102: Kanji-Hiragana conversion unit

103: Roman translator

104: synonym search unit

105: synonym recommendation

106: Answer word selector

Claims (27)

A roman conversion part for converting a pronunciation of a word expressed in Japanese hiragana form or katakana form into a roman character; And A similar word search unit for searching similar words for the word based on the converted roman characters Japanese automatic recommendation system that includes. The method of claim 1, The similar word search unit, Searching for a similar word for the word in consideration of a similarity score of the word converted to the roman character, The similarity score is, Japanese automatic recommendation, characterized in that it is determined based on at least one of the input frequency according to the length of the word, the edit distance according to whether the word is included in the long sound, midpoint, tactile sound or sound, or the degree of comparison of the circular state of the word system. The method of claim 2, The similar word search unit, And if the word is a Chinese character, the similarity score is determined in consideration of the comparison result of the form converted to Roman characters, the comparison result of the form converted to Hiragana, and the comparison result of the original form of the Chinese character. The method of claim 1, A similar word recommending unit for converting the searched similar words into a Japanese form of any one of the hiragana, katakana, and kanji. Japanese automatic recommendation system that includes more. The method of claim 4, wherein The analogous word recommendation unit, (1) if the difference between the similarity of the state converted to Roman characters and the similarity of the state not converted to Roman characters exceeds a preset criterion, the similar word is not recommended, or (2) Japanese automatic recommendation system, characterized in that the similar words are not recommended when the words are used more than the recommended similar words. The method of claim 4, wherein The analogous word recommendation unit, The automatic Japanese recommendation system, characterized in that for converting the searched analogous words in a form different from the Japanese form of the word. The method of claim 1, An error determination unit that analyzes the input word to determine whether the word is a typo More, The roman conversion unit, If the input word is a typo, Japanese automatic recommendation system, characterized in that for converting the word to Roman. The method of claim 7, wherein The typo determination unit, Whether the word is a typo in consideration of whether the word is included in preset typo data, whether the word input frequency or document appearance frequency is lower than a preset reference frequency, or whether the word is morphologically separated. Japanese automatic recommendation system, characterized by judging whether or not. The method of claim 7, wherein When the word is a typo, a correct word selection unit for selecting the correct word for the word among the searched similar words in consideration of the similarity score or the editing distance according to the frequency of input of the word Japanese automatic recommendation system that includes more. The method of claim 1, When the input word is a Chinese character, the Chinese character-hiragana conversion unit divides the word by token using token division learning data, and converts the word into hiragana corresponding to the divided token using Chinese character-Hiragana conversion learning data. Japanese automatic recommendation system that includes more. The method of claim 10, The token split learning data is, The Japanese automatic recommendation system, characterized in that it is determined through a hidden Markov Model (HMM) -based spacing learning using a corpus divided by the morpheme token of the Chinese character. The method of claim 10, The kanji-hiragana conversion learning data, A bigram dictionary and a unigram dictionary determined through learning based on a corpus separated by morpheme tokens of Chinese characters, The bigram dictionary is constructed with a frequency between tokens, The unigram dictionary is a Japanese automatic recommendation system, characterized in that the frequency is built between the token and Hiragana. The method of claim 12, The Kanji-Hiragana conversion unit, The Japanese automatic recommendation system, characterized in that for searching the bigram dictionary for the divided token to select a token representing the maximum probability, and converts the selected token to the hiragana corresponding to the unigram dictionary. Converting a pronunciation of a word expressed in Japanese hiragana form or katakana form into romaji; And Searching for similar words for the word based on the converted roman characters Japanese automatic recommendation method that includes. The method of claim 14, Searching for a similar word for the word, Characterized in that to search for a similar word for the word in consideration of the similarity score of the word converted to the roman alphabet, The similarity score is, Japanese automatic, characterized in that determined based on at least one of the frequency of input according to the length of the word, the edit distance according to whether the word contains a long sound, a midpoint, a tactile sound or the sound of the word or the degree of comparison of the circular state of the word Recommended way. The method of claim 15, Searching for a similar word for the word, And if the word is a Chinese character, the similarity score is determined in consideration of the comparison result of the form converted to Roman characters, the comparison result of the form converted to Hiragana, and the comparison result of the original form of the Chinese character. The method of claim 14, Recommending the searched analogous word by converting the searched analogous word into a Japanese form of the hiragana letter, katakana word or kanji character. Japanese automatic recommendation method that includes more. The method of claim 17, Recommend by converting the searched analogous words into any one of the Japanese form of the hiragana, katakana or kanji, (1) if the difference between the similarity of the state converted to Roman characters and the similarity of the state not converted to Roman characters exceeds a preset criterion, the similar word is not recommended, or (2) Japanese automatic recommendation method, characterized in that the similar words are not recommended if the words are used more than the recommended similar words. The method of claim 17, Recommend by converting the searched analogous words into any one of the Japanese form of the hiragana, katakana or kanji, The automatic Japanese recommendation method, characterized in that for converting the searched analogous words in a form different from the Japanese form of the word. The method of claim 14, Analyzing the input words to determine whether the words are typos More, Converting the pronunciation of the word to Roman, If the input word is a typo, the Japanese automatic recommendation method, characterized in that for converting the word to Roman. 21. The method of claim 20, Determining whether the word is a typo, Whether the word is a typo in consideration of whether the word is included in preset typo data, whether the word input frequency or document appearance frequency is lower than a preset reference frequency, or whether the word is morphologically separated. Japanese automatic recommendation method, characterized by judging whether or not. 21. The method of claim 20, If the word is a typo, selecting a correct answer word for the word among the searched similar words in consideration of a similarity score or an edit distance according to a frequency of input of the word; Japanese automatic recommendation method that includes more. The method of claim 14, When the input word is a Chinese character, dividing the word by token using token split learning data and converting the word into hiragana corresponding to the split token using the Chinese character hiragana conversion learning data. Japanese automatic recommendation method that includes more. 24. The method of claim 23, The token split learning data is, The Japanese automatic recommendation method characterized in that it is determined through the spacing learning based on hidden Markov model using the corpus divided by the morpheme token of the kanji. 24. The method of claim 23, The kanji-hiragana conversion learning data, Includes a bigram dictionary and a unigram dictionary determined through learning based on a corpus separated by morpheme tokens of Chinese characters, The bigram dictionary is constructed with a frequency between tokens, The unigram dictionary is a Japanese automatic recommendation method, characterized in that the frequency is built between the token and Hiragana. 24. The method of claim 23, Converting to the hiragana corresponding to the divided token, Searching a bigram dictionary for the divided tokens and selecting a token representing a maximum probability; And Converting to hiragana corresponding to a unigram dictionary for the selected token Japanese automatic recommendation method that includes. A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 14 to 26.

Images