KR20000045107A - Method for korean synonym search using phoneme level's appropriate function - Google Patents

Abstract

PURPOSE: A synonym search using phoneme level's appropriate function is disclosed to search for data without precise knowledge of data by searching in units of word, not in syllables, from ATTACH-based Directory Information System (ADAS), to examine the results in different steps, and to return the appropriate results in a numbered form. CONSTITUTION: A synonym search using phoneme level's appropriate function is composed of information search server, response device, and telephone. ADAS is an automatic telephone number information guide system that uses regular telephones(11) through public switched telephone network. As an input method, Korean code(ATTACH code) is inputted using telephone keypad, and as an output method an Automatic Response System device that displays voice or text is used. ATTACH code not only uses telephone buttons for input of Korean, English, special symbols, numbers, and control codes, it's also easily learned and mastered. Above search system is used by information search server(13) and information is output through voice response device(12) to the telephone(11).

Description

Hangul-like word retrieval method using phoneme fitness function

The present invention divides the number of cases in which a query word can be divided into phoneme units rather than syllables in an ATTACH-based Directory Information System (ADAS). The present invention relates to a method for retrieving similar words in Korean that enables users to search data even if they do not know exactly what data they are looking for by providing appropriate search results, and a computer-readable recording medium that records a program for realizing the same.

Data Retrieval compares words stored in a database to a query entered by a user and finds a match. However, data retrieval requires exact comparisons, so the user must know exactly what they are looking for.

Therefore, in the conventional data retrieval, the user can find the desired information without inputting all the long words, or search for the desired information by shortening each syllable or some syllables of the word that the user wants to input, or pressing the wrong button. There was a problem that can not support the desired information retrieval in case of occurrence.

In order to solve the problems described above, the present invention divides the number of cases in which the query word can be divided into phoneme units rather than syllables in the direct telephone number search system (ADAS), divided into several stages, and according to each stage, By providing the appropriate search results to users by assigning rankings, we provide a computer-readable recording medium that records Korean similar words for searching data and a program for realizing the data without knowing exactly what data to look for. There is this.

1 is an exemplary configuration of a direct telephone number search system (ADAS) to which the present invention is applied.

2 is a flowchart illustrating a method for retrieving Korean similar words using a fitness function of a phoneme unit according to the present invention.

* Explanation of symbols for the main parts of the drawings

11 telephone 12 voice answering device

13: information search server

According to the present invention for achieving the above object, in the Korean similar word search method applied to a telephone number search system, a search word comparison criterion is defined according to the number of cases in which a query word can be divided into phonemes instead of syllables. A first step of making; A second step of applying a weight to the defined search word comparison criteria to give priority to the search type; And a third step of, when the query word is input, searching for a partial match or a best match word by cutting the lower priority according to the priority.

The present invention also provides a telephone number retrieval system having a processor, the function of defining a search word comparison criterion according to the number of cases where a query word can be divided into phonemes instead of syllables; A function of applying a weight to the defined search word comparison criteria to give priority to the search type; And a computer-readable recording medium having recorded thereon a program for realizing a function of searching for a partial match or a best match by cutting out the lower priority according to the priority when the query word is input.

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

1 is an exemplary configuration of a direct telephone number search system (ADAS) to which the present invention is applied.

The telephone number direct search system (ADAS) to which the present invention is applied is a system for automatically guiding telephone number information using a general telephone 11 through a public switched telephone network (PSTN). 11) A Hangul code (ATTACH code) for inputting Hangul using the keypad of the present invention is used, and the output means is a terminal capable of displaying voice or text using an automatic response system (ARS) 12. Provide a text format.

ATTACH code is a very sophisticated interface that can input not only Korean characters, but also English, special symbols, numbers, and control codes as buttons on the phone 11. It can be easily learned and mastered.

However, even if the interface is well designed to input Korean and English with only twelve limited buttons, the user will feel burdened, and typing a long word will make a mistake.

In the telephone 11 character input environment, the user's character input should be made as short as possible and shortened input should be possible, and there is a need for a method for enhancing user input convenience such as automatically correcting an input error.

Accordingly, the function of the present invention was implemented in the information retrieval server 13 by introducing the concept of the input part in the design process of the direct telephone number search system, and the output is output to the telephone 11 through the voice response device 12.

Now, the Hangul word search method implemented in the direct phone number search system will be described in detail.

2 is a flowchart illustrating a method for retrieving Korean similar words using a fitness function of a phoneme unit according to the present invention.

The present invention basically corresponds to a data retrieval that finds a match by comparing a Hangul word (ie, a person's name, a company name, a business name, etc.) stored in a database to a query word input by a user.

However, data retrieval requires a complete match of the object, so the user must know exactly what to look for. Therefore, the commonly used data retrieval cannot meet the following requirements of the system.

First, users should be able to find the information they want without having to type all the long words.

Second, the user should be able to search for the desired information by inputting the syllables or some syllables of the words that the user wants to input, such as entering only consonants or only neutrals.

Third, even if an error occurs by pressing the wrong button, you need to support searching for the desired information.

Fourth, when searching through the above methods results in several matching objects, the search results should be provided in order of good matching properties. Here, since the search results are provided as voice, this requirement is very important as it requires a lot of audio output and listening time.

Fifth, even if a user does not exactly know a word to be searched, the user should be able to search for the desired information by inputting only a part of the word (for example, entering an abbreviation or the like). This requirement must be provided because the company name and business name registered in the telephone number database of the system may be different from those of general users, while the keyword index device is not designed in the database.

According to the present invention, when a proper result is not obtained in the conventional search method, the priority is given to each step by dividing the number of cases in which the query word can be divided into phonemes instead of syllables. Give them the appropriate search results so they can search the data without knowing exactly what they are looking for.

To establish the relevance function, set the following criteria.

First, the measurement for the substring is as follows.

In general, the concept of a substring used is that Q is regarded as a substring of T when the query Q is a continuous portion of the target string T. Here, when Q is also a substring to another object T ', a criterion is needed to determine which of T and T' is more suitable. Unfortunately, there can be no special criteria for any situation.

For example, when the user enters "school", it cannot be determined that either "university" or "high school" was intended. However, the following criteria can be set.

First, the shorter length of the part (ie, the syllable) except the matched part takes precedence. This criterion takes into account the voice output time, and if there are several objects that partially match the search results, the shorter one is outputted first because the user's time to find the target is reduced. In addition, the general logic that such a relative value of the amount of uncertain information is smaller than that of the other holds.

Secondly, the smaller the number of discordant syllables before the matched part, the higher the priority. This criterion also takes into account the audio output because it is easy to hear that the portion of interest is output first.

According to these first and second criteria, when the query word "school" is entered, "university" takes precedence over "high school" for both partial matches of "university" and "high school".

On the other hand, if you compare "school love" with "university," the first is "university," but the second is "school love."

However, as mentioned above, it cannot be determined which is right, so there is only a reasonable weight between these two criteria.

Extending the notion of general partial match, when query Q is part of the target string T, if Q is a substring of T ("consecutive" is missing), the fourth requirement above (ie The search results are presented in order of best match when multiple objects appear. For example, when the query is "traditional", "Korea Telecom" is also subject to partial agreement. In this case too, comparison criteria are needed.

Third, the length of the mismatched portion added to the partially matched portion takes precedence.

For example, when the query word is "traditional", "Korean communication" takes precedence over "Hannari communication". Here, "Korean barrel" is regarded as a partial match. Thus, this criterion does not conflict with the preferred second criterion as the number of mismatched syllables in front of the matched first criterion where the length of the part except the matched section prevails is preferred.

On the other hand, look at the criteria (Measure) for the shortcut input in detail as follows.

Let's look at the input method of the ATTACH code to develop a measure for the first requirement above (that is, the user should be able to find the desired information without having to type all the long words).

ATTACH code uses Hangul 2 beol type input method.

For example, "one" means " (Buttons 8 and 7) "+" (2 and 3) "+" (2 and 1) ".

Therefore, if the user enters "Korea" shortened, " "," Station "," study ", etc., where" "," Bottom "is a partial syllable of" one "," "," Gu "is considered to be a syllable of" station "and expanded to interpret" "," Station "," Estuary ", etc. can all be regarded as a substring of" Korea. "From this point of view, the concept of substring can be applied, that is, the following comparison criteria can be derived.

Fourth, a small syllable sum of the lengths of mismatched portions added to portions partially matched by syllable units takes precedence.

According to the criteria above, the query " "Hoku" takes precedence over "korea" because "ho" entered "han" rather than "han" Because the uncertainty for "is small. Also, the query" &Quot; Furniture " takes the priority over " powerhouse " when "

Next, the first requirement (i.e. the user should be able to find the information they want without having to type all the long words) and the fourth requirement (i.e. when the number of approximate matches come out is good). Ordering the search results), a criterion is needed to enter only the consonants for the purpose of prefixing.

For example, to search for "KT", Communication ", where" "What this query means in terms of this prefix" Of all "communications" (or "XXcommunications") beginning with " Search for "there's an X that starts with"). In this case, it is assumed that the first property of the prefix is one to reduce the complexity. For this case, the second criterion (i.e., the smaller the number of mismatched syllables in front of the matched part takes precedence) and the third criterion (i.e., the shorter length of the mismatched part added to the matched part takes precedence). By extension, new criteria can be derived:

Fifth, the smaller the number of discordant syllables in front of the syllable that corresponds to the prefix, the higher the priority. This criterion means that it is more intuitive to use the initial letter of the first letter of a word as a prefix.

Meanwhile, the criteria for error correction will be described in detail as follows.

Error correction is very difficult to develop a mechanism for obtaining reasonable and consistent results, so we will only deal with the case of neutral misinput, which is a common practice for users.

In the case of the first generation, since the proposed criterion plays a very important role, it is difficult to tolerate the initial error, and the finality can be ignored as long as the short input is provided.

A typical example of a neutral typing error is a spelling error.

For example, " "Wow " "Confused of" "Wow " ", Especially in foreign language notation. Thus, the following simple criteria can be set.

Sixth, the smaller the approximated neutral, the higher the priority.

For example, if "supermarket" is entered as a query, "supermarket" takes precedence over "supermarket." Once there is an error, this criterion is less ambiguous, but it has the advantage of presenting candidates for objects in the event of a possible misinput.

Now, the search function will be described in more detail as follows.

When comparing the search word comparison criteria from the first criterion to the sixth criterion as described above, it is shown in (Table 1).

Description of the Measure name standard The number of syllables except the matched part (first criterion) DIS_KEY The smaller the first Number of forward mismatched syllables except second match (second base) DIS_PRE Number of mismatched syllables added in discontinuously matched parts (third criterion) DIS_ISO Number of discrepancies added in discontinuous match parts in syllable units (fourth criterion) DIS_PAR Prefix position of the matched syllable (fifth) DIS_POS Approximate number of neutrals (sixth) DIS_APX

As mentioned above, some of these standards may conflict with each other and may be applied in duplicate. Thus, the search function takes a method of applying weights to the criteria between them. To do this, you first need to know and normalize the range of values that each criterion has.

Let Q be the query and T be the target word.

Here, when L (T) is the number of syllables of the word T, the name DIS_KEY of the first criterion (that is, the number of syllables except the matched portion) is an integer value from 0 to L (T) -L (Q). Have

The name DIS_PRE of the second criterion (ie, the number of forward mismatched syllables except for the matched part) also has an integer value from 0 to L (T) -L (Q).

Also, the discontinuous matched portion of the third criterion (ie, the number of disagreeable syllables added within the discontinuous matched portion) is up to L (T), so DIS_ISO is also 0 to maximum L (T) -L. (Q) up to.

In addition, if the number of the letters in the syllable t is l (t) in the fourth criterion (i.e., the number of inconsistent letters added in the discontinuous matched part by syllable unit), DIS_PAR is at most (q (Q (l (Q It has an integer value up to) -1).

Also, the name DIS_POS of the fifth criterion (the position of the prefix-matched syllable) has a value from 0 to L (Q) -2 when 0 is the first syllable.

In addition, the name DIS_APX of the sixth criterion (ie, the approximate number of neutrals) ranges from 0 to a maximum of L (Q).

As described above, if the normalization is performed based on the minimum maximum value, the range of values taken by each criterion can be set from 0 to 1.0.

When the normalized value of each criterion is expressed as N (...), the relevance function is expressed as follows. In other words, assuming a set of criteria, M = {DIS_KEY, DIS_PRE, DIS_ISO, DIS_POS, DIS_APX} and assuming that the weight for criterion m is W (m) (a real value between 0 and 1), the search = { 1-(m (M, W (m)) (N (m)) x 100%.

The above function has a percentage value from 0 to 100 and is 100% if Q matches T completely.

Finally, when there are many search results for the query, it is necessary to cut to a certain number or less.

Cutting is not important in general IR, but it is advantageous in the telephone 11 voice / text information service that takes a long time to reduce the search target. In addition, cutting may be a method of discarding an object whose search value is equal to or less than a predetermined level, but a more effective method may be to cut a lower priority by giving priority to criteria. For this purpose, the search types are classified as shown in Table 2.

type Explanation Priority RV_EXACT Exact match One RV_P_KEY Initial prefix + substring 2 RV_KEY Substring 3 RV_P_PAR Initial prefix + syllable unit substring 4 RV_PAR Syllable Unit Substring 5 RV_P_APX Initial Prefix + Neutral Approximation 6 RV_APX Neutral approximation 7

Referring to (Table 1) and (Table 2), look at the effect of the criteria according to the search type, as shown in (Table 3).

As described above, the Korean similar word search method using the fitness function of the phoneme unit according to the present invention may be classified into phonemes instead of syllables so that the data can be searched even if the data to be searched is not known exactly. According to the number of search terms, search word comparison criteria are defined (see Table 1), and weights are applied to the defined search word comparison criteria, and priority is given according to the search type (see Table 2).

Then, when the word to be searched is input (301), it is searched from the database (302), exact match (RV_EXACT), initial prefix + substring (RV_P_KEY), substring (RV_KEY), and consonant according to the priority according to the criterion type. The prefix + syllable unit substring (RV_P_PAR), the syllable unit substring (RV_PAR), the initial prefix + neutral approximation (RV_P_APX), and the neutral approximation (RV_APX) are sequentially examined (203 to 208).

Finally, the search results searched according to the priority are output (209).

As described above, the present invention is a word search method that selects partial matches or best matches, not exact matches, and is less civilized in error, and information or terms that the user knows in a situation where it is difficult to memorize accurate terms and information. It provides users with access to the results they want by using only the information and communication industry such as searching the database and searching the words stored in the file, correcting the spelling in word processing, and when the input of words such as the phone or mobile terminal is restricted. It can be utilized.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the spirit of the present invention for those skilled in the art to which the present invention pertains, and the above-described embodiments and accompanying It is not limited to the drawing.

As described above, the present invention divides the number of cases in which the query word can be divided into phoneme units instead of syllables in a direct telephone number search system (ADAS), and divides the number of cases into several stages to give priority to each step, so as to be appropriate for the user. By providing a search result, an appropriate result can be obtained even if the user does not know the word correctly, cannot input all the words, or if the input is restricted.

Claims (4)

In the Korean similar word search method applied to the phone number search system, A first step of defining a search word comparison criterion according to the number of cases in which the query word can be classified by phoneme rather than syllable; A second step of applying a weight to the defined search word comparison criteria to give priority to the search type; And A third step of searching for a partial or best match word by cutting out the lower priority according to the priority when the query word is input; Korean similar word retrieval method using the goodness-of-fit function of the phoneme unit, including. The method of claim 1, The first step is, A fourth step of prioritizing that the length of the portion (syllable) except the matched portion is short for the substring; A fifth step of prioritizing for the substring the smaller the number of discordant syllables before the matched part; A sixth step of prioritizing that the length of the mismatched portion added to the partially matched portion is short for the substring; A seventh step of prioritizing that the sum of the syllables of the lengths of the mismatched portions added to the portions matched by the digit unit is small for the short-term input; An eighth step of prioritizing the shorter input, the smaller the number of discordant syllables in front of the syllable corresponding to the prefix; And The ninth step, with less approximation of neutrality, prioritizing for error correction Korean similar word retrieval method using the goodness-of-fit function of the phoneme unit, including. The method according to claim 1 or 2, With respect to the search word comparison criteria defined above, by knowing and normalizing the range of values of each criterion and applying the weights between the criteria, perfect match (RV_EXACT), initial prefix + substring (RV_P_KEY), substring (RV_KEY) Goodness of phoneme units characterized by priorities in order of first prefix, syllable unit substring (RV_P_PAR), syllable unit substring (RV_PAR), initial prefix + neutral approximation (RV_P_APX), and neutral approximation (RV_APX) How to search Korean similar words using functions. In a telephone number search system having a processor, A function of defining a search word comparison criterion according to the number of cases in which the query word can be classified by phoneme instead of syllables; A function of applying a weight to the defined search word comparison criteria to give priority to the search type; And When a query is input, a function of searching for a partial or best match word by cutting out the lower priority according to the priority A computer-readable recording medium having recorded thereon a program for realizing this.