KR19990053161A - Search Method of Related Verbs for Semantic Information Retrieval - Google Patents

Abstract

The present invention can automatically extract relevant verbs used to resolve semantic ambiguity between nouns from a large amount of corpus to actively exclude human subjectivity, and relate to semantic-based information retrieval that can build highly reliable information. It is about how to search for verbs.

Recently, in the field of information retrieval, Hangeul natural language processing technology has been able to detect the parts of speech and its original form and its role in sentences through morphological analysis and syntax analysis. However, the information of the sentence itself is not enough to raise the accuracy and reproducibility of the search to the level that humans can think of. In addition, this information is used to resolve semantic ambiguity between nouns in application systems in the natural language processing field (thesaurus, semantic network, etc.), but there are many problems in constructing such information. Until now, information on related verbs has been built purely by human hand, which is related to the general level of common sense of the person who builds them.

In order to solve the above problems, the present invention automatically extracts relevant verbs used to solve semantic ambiguity between nouns from a large amount of corpus to actively exclude subjectivity of a person, and can construct highly reliable information. Suggest a search method.

Description

Relevant Verb Search Method for Semantic Information Retrieval

The present invention can automatically extract relevant verbs used to resolve semantic ambiguity between nouns from a large amount of corpus to actively exclude human subjectivity, and relate to semantic-based information retrieval that can build highly reliable information. It is about how to search for verbs.

Recently, in the field of information retrieval, Hangeul natural language processing technology has been able to detect the parts of speech and its original form and its role in sentences through morphological analysis and syntax analysis. However, the information of the sentence itself is not enough to raise the accuracy and reproducibility of the search to the level that humans can think of. In the light of the fact that the elements in the sentence that have the greatest influence on the results of IR, the noun extension to improve the accuracy and reproducibility of the search can overcome the limitations of simple search based on pattern matching. It is an important means.

Considering the characteristics of the language as described above, it can be said that the verb group or adjective group used with this noun in a sentence composed of a noun with a specific meaning has a certain rule with each other. For example, such as "Cap ^1" ( ^"verbs that can be used with this when you view a noun that writing to the head of the person) are 'bitter' or 'lay off'. The verb that is used with the noun 'mother ² ' (mother and son) may be of the kind 'good' or 'bad' between mothers and mothers, but it is never a group of verbs such as 'write' or 'naked'. will be. That is, nouns and verbs (verbs, adjectives) have information that can accurately grasp the meaning of each other.

This information is used to resolve semantic ambiguity between nouns in natural language processing system. The related verb searcher for semantic-based information retrieval is used for a language processing problem that may occur when a user wants to search for desired information among files stored in a computer, a document stored in a library and an electronic library, and the Internet. In particular, to resolve ambiguity between words, it is used to construct a semantic dictionary based on semantics.

The prior art has solved the above problems by constructing a thesaurus or semantic network with much effort to solve the above problems. However, these could only be written manually through the human head and hands.

Therefore, information construction for a large number of words had to be done by many people and was time consuming. This required effort and expense to produce a lot of error information and manage it.

Therefore, in order to solve the above-mentioned problems, the present invention expands a query through a noun dictionary automatically constructed by applying a relevant verb automatic extraction algorithm from a large amount of corpus, thereby increasing the reproducibility of the search and the meaning of the word appearing in the sentence. The purpose is to resolve ambiguities.

According to an aspect of the present invention, after a sentence and a word are input, a first step of extracting a word from the input sentence and confirming whether the sentence is at the end, and if the result is not the end of the sentence, In the second step of storing the result of the morpheme analysis of the word in the buffer, and confirms the existence of the noun, if the noun does not exist, return to the first step to extract the word from the sentence A third step of setting a found flag when the noun exists; a fourth step of confirming whether a found flag exists at the end of the sentence as a result of the checking of the first step; If is not present, terminate. If there is the found flag, extract one or more candidate words from the stem buffer, and then pattern 1, pattern 2 and pattern 3. A fifth step of checking whether or not it is related to pattern recognition, and if it is not related, returns to the fifth step and performs a repeated operation until there is a related pattern, and if there is a related pattern, Storing the result as a verb, and checking the end of the buffer; and if the end of the buffer is the end of the buffer, returns to the fifth step if the buffer is not the end, and extracts candidate words from the morpheme buffer. And a seventh step of repeating the operation.

1 is a hardware configuration diagram applied to the present invention.

2 is a module configuration diagram of a system applied to the present invention.

3 is a flowchart illustrating a method of searching for related verbs according to the present invention;

<Description of Signs of Major Parts of Drawings>

11: Processor (Pi (1 <= i = n)) 12: Main memory memory area (Mem)

13: shared memory area (SM) in memory

14: Input / Output Dedicated Processor (IOP) 15: Auxiliary Memory (DISK)

201: Related verb searcher 202: Corpus searcher

203: stemmer searcher 204: dictionary builder

205: noun dictionary 206: XDBM interface

The general content of the related verb extraction required in the present invention will be described, and the definition of the related verb is as follows.

"Relevant verbs are defined as related verbs that are semantically related to the relevant noun. Relevant verbs consist of a prototype group of verbs and adjectives."

Related verbs refer to a verb that is semantically related to the noun. For example, in the sentence "read a book," the related verb for "book" becomes "read." The form of related verbs can be thought of as three patterns.

In [Pattern 1], the nouns that make up the subject or object in the case of non-minist sentences are generally considered to be semantically related to the verbs that constitute the sentence.

For example, "baby sleeps" means "baby (the noun)"-"sleep" (related verb), and "student reads the book" "student (the noun)"-"read (related verb)" and "book." (Corresponding noun) '-' read (related verb) 'is regarded as related.

In addition, depending on the form of the sentence, the relationship between the middle-verb or the object-verb varies.

For example, in "students read books," there are many verbs that can be related to the noun 'student' and relatively few verbs that can be related to the noun 'book'.

In [Pattern 2], in the case of a ministry sentence, a noun forming a lexical adverb, an action verb, and a verb of a sentence have a practical semantic relationship, and the subject of the sentence represents only a ministry relation.

If the ministry statement that targets dative busaeo on, "the teacher was read a book to a student (he reads)" deep structure of the teachers [+ Ministry] _statement [student read a book] _statement "students (those in the sentence Noun) '-' read (related verb) 'and' book (the relevant noun) '-' read (related verb) 'has a semantic relationship.

In the case of a ministry statement for an object, the in-depth structure of "Cheol melted candles" is "cho" (noun)-"melt" (related verbs) ) Has a semantic relationship.

In [Pattern 3], a tubular word consisting of a tubular ending attached to a stem of a verb and a noun receiving a modification of the tubular word have a semantic relation. Since the prototype of a noun that modifies the noun is a verb or adjective, they become associated verbs.

For example, "a book that has been read a lot" can be called "the book has been read a lot," and a book has a semantic relationship.

The three patterns described above are the basis for automatic extraction of related verbs.

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

1 is a hardware configuration diagram applied to the present invention, showing the environment (environment) of the operating hardware.

Pi (1 <= i = n) 11 is a processor, Mem 12 is a main memory memory area, SM 13 is a shared memory area in memory, and IOP 14 is an I / O dedicated processor; DISK 15 is a disk which is an auxiliary storage device. As such, the present invention may operate in a general hardware environment including a single processor or multiple processors and a shared memory area.

2 is a configuration diagram showing a software module of the system to which the present invention is applied, and shows a configuration diagram of a noun dictionary builder to which a related noun searcher is applied.

The related verb searcher 201 inputs a sentence from the corpus searcher 202 and recognizes a pattern from the sentence with the help of the morphological analyzer 203 and registers the relevant noun registered in the noun dictionary 205 through the dictionary builder 204. Enter the relevant verb. The interface between the system and the corpus / dictionary is dedicated to the XDBM interface module 206.

3 is a flowchart illustrating an operation of searching for a related verb according to the present invention, and recognizes a pattern related to the above-mentioned noun from an input sentence and extracts a term, ie a verb or a tubular word, in the pattern.

The operation flowchart of the related verb searcher can be broadly classified into two processes. First, the process of determining whether there is a corresponding noun from the input sentence, and second, if the noun exists, it can be considered as a process of recognizing the above-described pattern from the result of the morphological analysis.

One word from the input sentence and the word (corresponding noun) 301 is extracted from the sentence (302). The extracted words are analyzed with the help of a morpheme analyzer, and the results are stored in buffers (303 and 304). This process is applied to all words in a sentence. If the relevant noun is found in the sentence (305), the flag is set to be known in the next step (306).

After morphological analysis of all the words for a character, and if the noun is found (307), one or several candidate words are extracted from the morphological analysis buffer (308). This process proceeds by considering the window size (that is, the number of words) as much as a predefined number in the pattern recognition process performed next. When the extracted word is found to be a related verb by pattern recognition (309), it is stored as a related verb (310). The above process is performed for all words stored in the morphological analysis buffer (311).

As described above, according to the present invention, the related verb dictionary constructed by the related verb automatic extraction algorithm for semantic-based information retrieval has the following excellent effects.

First, when information such as a document desired by a user is not extracted during information retrieval, it is possible to provide more information to the user by applying to the expansion of a query to increase the reproducibility.

Second, the ambiguity of the meaning of the multiword word that can occur in the process of analyzing natural language sentences can be identified and solved.

Third, the basis for classifying a lot of information by field is to obtain a set of words representing each field. Through this, it is possible to distinguish which field the input documents are. This document classification can also be solved using information from related verbs.

Fourth, in providing the reader with information such as daily newspapers, the reader can select only the desired information and provide the technology to determine whether the currently produced information is the desired information. This can be solved using relevant verbs.

Fifth, technically advanced dictionary construction is possible by expanding the parts that can be automatically constructed in constructing semantic dictionaries such as thesaurus and semantic networks.

Claims (2)

After a sentence and a word are input, extracting a word from the input sentence and checking whether the sentence is at the end; If the result of the check is not the end of the sentence, storing the result of the morphological analysis of the word in a buffer, and then checking whether the noun exists; A third step of returning to the first step if the noun does not exist, extracting a word from the sentence, and setting the found flag if the noun exists; A fourth step of checking whether a found flag is present at the end of the sentence as a result of the checking of the first step; As a result of the check, if the found flag does not exist, the process is terminated. If the found flag exists, one or more candidate words are extracted from the morpheme buffer, and then the pattern recognition of the pattern 1, pattern 2, and pattern 3 is performed. A fifth step to check whether If the result of the check is not related, the method returns to the fifth step and performs the repeated operation until the related pattern exists, and if the related pattern exists, stores the related verb, and then confirms whether or not the buffer ends. With six steps, And a seventh step of repeating an operation of extracting candidate words from the morpheme buffer by returning to the fifth step if the end of the buffer ends. How to retrieve related verbs for information retrieval. The method of claim 1, wherein the pattern recognition is In the case of non-mineral sentences, the nouns that make up the subject or object have a pattern 1 that is considered to be semantically related to the verbs that make up the sentence, In the case of the ministry sentence, pattern 2, which considers that the nouns that form the marginal adverb and the object and the verb of the sentence have a practical semantic relationship, and the subject of the sentence represents only the ministry relation, A method for searching related verbs for semantic-based information retrieval, comprising: a tubular word consisting of a tubular ending attached to a stem of a verb and a noun receiving a modifier of the tubular word, including a pattern 3 which is considered to have a semantic relation.