Patent Application
20180267957
This application claims priority under 35 U.S.C. § 119 to Korean Patent Application No. 10-2017-0031766, filed on Mar. 14, 2017, the disclosure of which is incorporated herein by reference in its entirety.
The present invention relates to an apparatus and method of processing an online contextual customized advertisement based on a morpheme analysis result.
A morpheme denotes a minimum unit having a meaning in linguistics, and a morphological analyzer performs a function of analyzing a text by a morpheme unit the most suitable for a context. Generally, the morphological analyzer may be divided into a method based on a rule and a dictionary and a method based on machine learning.
In relevant technology, “probabilistic segmentation and tagging of unknown words (Bogyum Kim, Jae Sung Lee, 2016)” has proposed a method of segmenting and tagging coined words in the 3-step probabilistic morphological analysis. In detail, a segmentation and tagging method for unknown Korean words has been proposed for the 3-step probabilistic morphological analysis. For guessing unknown word, it uses rich suffixes that are attached to open class words, such as general nouns and proper nouns. The inventors have proposed a method to learn the suffix patterns from a morpheme tagged corpus, and calculate their probabilities for unknown open word segmentation and tagging in the probabilistic morphological analysis model. In such a method, a coined word pattern is learned and is combined with a conventional morpheme tagging model to increase a tagging performance for coined words, but an adverse effect occurs in general documents, causing the reduction in performance.
In another relevant technology, U.S. Pat. No. 8,275,607 (Title of the Invention: semi-supervised part-of-speech tagging) has proposed a method that allocates a part of speech to each word based on dictionaries, calculates a Baysian probability value of words unlisted in dictionaries by using surrounding context information as attributes, and allocates the most suitable part of speech. However, the method needs a dictionary and a learning set established through a manual process, and for this reason, if the field is changed, performance is reduced.
Accordingly, the present invention provides an apparatus and a method, which automatically extract a coined word having a high probability that abnormal analysis is performed when analyzing a morpheme, improve a morpheme analysis performance based on the extracted coined word, and improve online context customized advertisement technology by using an improved morphological analyzer.
The present invention also provides an apparatus and a method, which perform a morpheme analysis-based automatic verification for extracting only coined words, which cause an actual morpheme analysis error, from among extracted coined words, perform a manual verification on a coined word which has passed the automatic verification, and improve context customized advertisement technology of the advertisement field, where coined words appear frequently, by using a morphological analyzer which is obtained by improving a performance of each of a coined word extractor and a morphological analyzer based on a result of the manual verification.
The objects of the present invention are not limited to the aforesaid, but other objects not described herein will be clearly understood by those skilled in the art from descriptions below.
In one general aspect, a language analysis apparatus for processing coined word recognition includes: a document collector collecting an online document; a coined word extractor performing coined word filtering on the collected document based on a predetermined coined word extraction pattern and a dictionary to extract a coined word candidate which corresponds to the coined word extraction pattern and is not registered in a coined stop word dictionary; a coined word verifier performing morpheme analysis on the extracted coined word candidate to verify a coined word candidate which does not correspond to a predetermined part of speech and is not registered in a pre-stored morpheme dictionary; a coined word extraction improver refining the coined word extraction pattern, based on a verification result of the coined word verifier and updating the verified coined word candidate in the coined stop word dictionary; and a morpheme analysis improver updating the verified coined word candidate in the morpheme dictionary, based on the verification result of the coined word verifier.
In another general aspect, a contextual advertisement intellectualization apparatus including the language analysis apparatus includes: a document collector collecting an online document which is to be advertised; a language analyzer performing morpheme analysis on the collected online document by using the language analysis apparatus; a keyword extractor extracting at least one keyword, based on a result of the morpheme analysis; an advertisement matcher extracting advertisement targets matching the extracted at least one keyword from among registered advertisement targets; and an advertisement recommender recommending an advertisement corresponding to an advertisement target, having a highest correlation with the extracted at least one keyword, from among the extracted advertisement targets.
In another general aspect, a language analysis method based on a language analysis apparatus for processing coined word recognition includes: collecting an online document; performing coined word filtering on the collected document based on a predetermined coined word extraction pattern and a dictionary to extract a coined word candidate which corresponds to the coined word extraction pattern and is not registered in a coined stop word dictionary; performing morpheme analysis on the extracted coined word candidate to verify a coined word candidate which does not correspond to a predetermined part of speech and is not registered in a pre-stored morpheme dictionary; refining the coined word extraction pattern, based on a result of the verification and updating the verified coined word candidate in the coined stop word dictionary; and updating the verified coined word candidate in the morpheme dictionary, based on the result of the verification.
In another general aspect, a contextual advertisement intellectualization method based on a language analysis method of processing coined word recognition includes: collecting an online document which is to be advertised; performing morpheme analysis on the collected online document, based on the processing of the coined word recognition; extracting at least one keyword, based on a result of the morpheme analysis; extracting advertisement targets matching the extracted at least one keyword from among registered advertisement targets; and recommending an advertisement corresponding to an advertisement target, having a highest correlation with the extracted at least one keyword, from among the extracted advertisement targets, wherein the language analysis method includes: collecting an online document; performing coined word filtering on the collected document based on a predetermined coined word extraction pattern and a dictionary to extract a coined word candidate which corresponds to the coined word extraction pattern and is not registered in a coined stop word dictionary; performing morpheme analysis on the extracted coined word candidate to verify a coined word candidate which does not correspond to a predetermined part of speech and is not registered in a pre-stored morpheme dictionary; refining the coined word extraction pattern, based on a result of the verification and updating the verified coined word candidate in the coined stop word dictionary; and updating the verified coined word candidate in the morpheme dictionary, based on the result of the verification.
Other features and aspects will be apparent from the following detailed description, the drawings, and the claims.
Hereinafter, embodiments of the present invention will be described in detail to be easily embodied by those skilled in the art with reference to the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. In the accompanying drawings, a portion irrelevant to a description of the present invention will be omitted for clarity. Like reference numerals refer to like elements throughout. Also, in providing description with reference to the drawings, although elements are represented by the same name, reference numeral referring to the elements may be changed, and reference numerals are merely described for convenience of description. It should not be construed that concepts, features, functions, or effects of elements are limited by reference numerals.
In this disclosure below, when it is described that one comprises (or includes or has) some elements, it should be understood that it may comprise (or include or has) only those elements, or it may comprise (or include or have) other elements as well as those elements if there is no specific limitation. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. The meaning of ‘comprise’, ‘include’, or ‘have’ specifies a property, a region, a fixed number, a step, a process, an element and/or a component but does not exclude other properties, regions, fixed numbers, steps, processes, elements and/or components.
In the present specification, each of terms such as “ . . . unit”, “ . . . apparatus” and “module” described in specification denotes an element for performing at least one function or operation, and may be implemented in hardware, software or the combination of hardware and software.
Hereinafter, a language analysis apparatus which improved morpheme analysis performance based on automatic coined word extraction according to an embodiment of the present invention will be described in detail with reference to
First, as illustrated in
The document collector 110 may collect a target document which coined words are to be extracted from.
A document collection target, a collection method, and collection technology of the document collector 110 are not limited to a specific target, method, and technology. For example, the document collector 110 may randomly collect sentences from online text documents, and moreover, may collect sentences which are input as queries by an arbitrary user.
The coined word extractor 120 may extract a coined word appearing in sentences which are results of the document collection.
Referring to
An operation and a process which are performed in the pattern-based coined word extraction operation are as follows.
A pattern for extracting coined words may be configured based on “surrounding words” such as a postposition and an ending which are used along with the coined words. Such a coined word extraction pattern may be configured in various methods, and is not limited to one method.
In an embodiment of the present invention, it is assumed that a noun coined word is extracted based on “postposition”. Since most of nouns are used along with postpositions, the other parts except a part capable of becoming a postposition from an end in each syntactic word may be used as nouns and may be assumed as coined word candidates.
For example, when a sentence “” is collected, a list of “postpositions” may include “
,
”. A result obtained by extracting syntactic word-based coined words based on the list is shown in the following Table 1.
In the sentence, ‘’, ‘
’, ‘
’, and ‘
’ may be extracted as the coined word candidates.
A coined word pattern database (DB) 121 applied to the pattern-based coined word extraction operation may be included in the coined word extractor 120, but may be provided in a separate storage space of the language analysis apparatus 100 without being limited thereto.
An operation and a process which are performed in the dictionary-based filtering operation will be described below.
In a dictionary-based coined word extraction operation, the coined word candidates extracted in the pattern-based coined word extraction operation may be filtered based on a coined stop word dictionary. In this case, a stop word dictionary may be a list which is extracted as a noun although not a noun. For example, when it is assumed that ‘’ is not included in stop words included in the stop word dictionary, any one of the coined word candidates is not filtered in the dictionary-based filtering operation. Therefore, ‘
’, ‘
’, ‘
’, and ‘
’ may be selected as final coined word candidates.
A coined stop word database 122 (illustrated as “coined stop word 1 database” in
Returning again to
First, in the dictionary-based automatic verification operation, coined words extracted by the coined word extractor 120 may be filtered based on a morpheme dictionary 131 (illustrated as “morpheme dictionary 1 database” in ’ is registered in the morpheme dictionary, ‘
’ may be excluded from the coined word candidates, and only ‘
’, ‘
’ and ‘
’ may be left.
Subsequently, in the morpheme analysis-based automatic verification operation, verification may be automatically performed based on a morphological analyzer 132 (illustrated as “morpheme model database” in
For example, it is assumed that a morpheme analysis result for an exercise sentence from which a coined word candidate is extracted is as follows. That is, when the exercise sentence is “”, a morpheme analysis result for the exercise sentence is shown in the following Table 2.
/noun +
/noun +
/vcp +
/etn +
/jks
/noun +
/xsv +
/etm
/noun +
/jkg
/noun
/noun +
/noun
/noun +
/jkb
/noun +
/xsv +
/ep +
/ef + ./s
In defining a part of speech, noun, vcp (predicative postposition), etn (noun changing ending), jks (nominative postposition), xsv (deverbative suffix), etm (prenoun changing ending), jkg (genitive postposition), jkb (adverbial postposition), ep (prefinal ending), ef (final ending), and s (symbol) may be listed in Table 2.
Moreover, an automatic verification result for each of coined word candidates is shown in the following Table 3.
/noun +
/noun +
/vcp +
/etn +
/jks
/noun +
/noun
/noun +
/jkb
As shown in Table 3, ‘ ’ and ‘
’ of the coined word candidates are not analyzed as a noun, and thus, are determined as verification fail. In the coined word candidates, ‘
’ is analyzed as a noun, and thus, is determined as verification pass.
Finally, in the manual verification operation, coined word candidates which are determined as fail in the automatic verification operations may be manually verified. In detail, when a coined word candidate is analyzed as a noun as a result of morpheme analysis and thus is determined as verification pass, morpheme analysis and coined word extraction may not secondarily be performed, and the coined word candidate may be excluded from manual verification. On the other hand, when coined word candidates are determined as fail in automatic verification, this may be classified into two cases as shown in the following Table 4, and the two cases may be manually verified according to different processing methods.
For example, ‘’ of coined word candidates may correspond to the case 1 and thus may be stored in a coined word dictionary database 134, and ‘
’ may correspond to the case 2 and thus may be stored in a coined stop word dictionary 133 (illustrated as “coined stop word 2 database” in
Returning again to
In detail, referring to
First, the coined stop word-based coined word pattern refining operation may process a case corresponding to the case 2 in the manual verification operation. That is, when the coined word extractor 120 extracts a noun coined word but the noun coined word is not a noun, the extracted noun coined word may be stored in the coined stop word dictionary 133 (i.e., the “coined stop word 2 database”). Therefore, based on the ‘’ stored in the coined stop word dictionary 133, the coined stop word-based coined word pattern refining operation may refine a relevant pattern so as not to extract a word having a similar type as a coined word candidate. At this time, the coined word extraction improver 140 may correct the coined word pattern database 121, which is a previous coined word extraction pattern dictionary, of the coined word extractor 120 to effect an update.
Subsequently, the coined stop word dictionary integration operation may be an operation of integrating a coined stop word dictionary for excluding coined words, which are not a noun coined word, from the coined word extractor 120. In the coined stop word dictionary integration operation, content of a new coined stop word dictionary 133 (i.e., the coined stop word 2 database) may be added to the coined word pattern database 121, which is a previous coined word extraction pattern dictionary, of the coined word extractor 120. For example, since the ‘’ is extracted as a new coined stop word, the ‘
’ may be added to the coined stop word 1 database which is a previous dictionary.
Returning again to
Referring to
First, in the morpheme dictionary generation operation, a processing target may be a case corresponding to the case 1 in a manual operation performed by the coined word verifier 130. That is, the case may be a case where a coined word candidate is a noun actually but is not analyzed as a noun by the morphological analyzer, the coined word candidate may be stored in a coined word dictionary in the manual operation. In the morpheme dictionary generation operation, ‘’ which is stored in the coined word dictionary 134 in the manual operation may be generated as a morpheme dictionary 131 (i.e., the morpheme dictionary 1 database). In this case, the morpheme dictionary 131 may be a dictionary applied to the morphological analyzer, and registered words may be used for analyzing morphemes.
A format of the morpheme dictionary 1131 may be diversified depending on the morphological analyzer, and is not limited to one type.
For example, in the manual operation performed by the coined word verifier 130, a target corresponding to the case 1 may be stored in the morpheme dictionary 1131 in order for a word to match a correspond part of speech as in the following Table 5.
Subsequently, in the morpheme dictionary integration operation, a newly generated morpheme dictionary 151 (illustrated as “morpheme dictionary 2 database in
NOUN
NOUN
NOUN
The morpheme dictionary 2151 generated through the above-described operation may be used in the morphological analyzer, and thus, a previously misanalysed sentence may be accurately analyzed as follows. That is, a syntactic word “” may be analyzed as “
/noun+
” in a collected sentence.
The language analysis apparatus 100 according to an embodiment of the present invention solves a problem where a morpheme of a coined word which is not included in a dictionary or learning data is not accurately analyzed, and thus, may be applied to the advertisement field where a number of coined words appear. Contextual customized advertisement technology which is significant technology in the advertisement field may be technology for recommending an advertisement suitable for context of a body text such as news articles, and in the contextual customized advertisement technology, morpheme analysis technology may perform an essential function. Therefore, the language analysis apparatus 100 according to an embodiment of the present invention improves a customized advertisement recommending function by using the contextual customized advertisement technology.
Hereinafter, a contextual advertisement intellectualization apparatus and an intellectualization service method thereof according to an embodiment of the present invention will be described in detail with reference to
In
First, a document may be collected in step S610.
For example, the collected document may be a part of a news article associated with vehicles, and a sentence “” may be a processing target.
Subsequently, language analysis may be performed in step S620.
In this case, as a processing result obtained in step S620, a result obtained by performing morpheme analysis on content of the collected document may be shown in the following Table 7.
:NP>
<
|
:NNG>
:NNG> <
:NNG>
:NNG> <
:NNG>
<
:NNG>
.
That is, “” in “
” may be analyzed as a pronoun. For reference, in Table 7, only an example of each of a general noun (NNG) and a pronoun (NP) is shown, and a morpheme analysis result is not limited thereto. In other embodiments, classifications applied to general morpheme analysis may all be used.
Subsequently, keyword extraction may be performed based on a result of the language analysis in step S630.
In a keyword extraction operation, an essential keyword for recommending an advertisement suitable for content of the document may be extracted, and a noun may be extracted as the essential keyword.
As a processing result obtained in step S630, a result of the keyword extraction may be shown in the following Table 8.
Moreover, keywords associate with an advertisement target which is to be advertised by an advertisement provider may be previously stored in an advertisement database, and in step S640, advertisement matching suitable for the document content among advertisement targets registered in the advertisement database may be performed based on the keyword extraction result.
For example, various advertisement targets registered by vehicle-related advertisement providers may be stored in the advertisement database. In ” are stored in association with “
” is illustrated.
Subsequently, advertisement recommendation may be performed based on a result of the advertisement matching in step S650.
The advertisement recommendation operation may calculate a suitable level matching the document content, based on keyword-based priorities of advertisements selected through the advertisement matching and may finally recommend an advertisement the most suitable for the document content. However, a significant advertisement target keyword may not be included in keywords extracted based on an analysis result which is obtained in step S620. That is, in a collected sentence content, ‘’ which is a vehicle model may be a significant keyword, but a pronoun ‘
’ may be analyzed through morpheme analysis in the language analysis processing operation, whereby the morpheme analysis may be a case where the advertisement matching fails.
In order to solve such a problem, the contextual advertisement intellectualization apparatus according to an embodiment of the present invention may include a language analyzer corresponding to the language analysis apparatus 100 described above with reference to
As illustrated in
That is, as illustrated in
In ” which is not processed by conventional contextual customized advertisement technology is stored as a coined word in a morpheme dictionary in the sentence “
” collected in
The document collector 710 may collect various documents. To describe an operation of performing improved language analysis on ‘’, an example of collecting a document where ‘
’ appears will be described below.
The document collector 721 of the language analyzer 720 may previously collect various documents and may provide the collected documents to the coined word extractor 722. For example, the document collector 721 may collect a sentence “” and a sentence “
” and may provide the collected documents to the coined word extractor 722.
Therefore, the coined word extractor 721 may extract ‘’, and ‘
’ as coined words, based on postpositions ‘
’.
Moreover, the coined word verifier 723 may perform automatic verification and manual verification through dictionary and morpheme analysis to remove ‘’ from a coined word candidate and may maintain only ‘
’ and ‘
’ as coined words.
Subsequently, the coined word extraction improver 724 may register ‘’, removed by the coined word verifier 723, in a stop word dictionary to improve a performance of the coined word extractor 722. Also, the morpheme analysis improver 725 may register ‘
’ and ‘
’, which have passed verification by the coined word verifier 723, as nouns in a morpheme dictionary.
Therefore, the language analyzer 720 of the contextual advertisement intellectualization apparatus 700 of ’ in the collected sentence by using a performance-improved morpheme analyzer.
:NNG>
:NNG>
:NNG> <
:NNG>
:NNG> <
:NNG>
:NNG>
.
Furthermore, the keyword extractor 730 may extract keywords as shown in the following Table 10.
Subsequently, the advertisement matcher 740 may extract ‘’ including ‘
’ as an advertisement candidate from among various advertisement targets registered in the advertisement database 750, based on the extracted keywords.
Then, the advertisement recommender 760 may finally recommend ‘’ as an advertisement suitable for the document from among the recommended advertisement candidates.
First, an online document which a relevant advertisement is to be inserted in may be collected in step S910.
Subsequently, in step S920, language analysis based on automatic coined word recognition may be performed on content of the collected online document to perform morpheme analysis.
In this case, a language analysis processing method based on automatic coined word recognition may first collect various online documents in step S921, extract new appearing coned word candidates from the collected documents in step S922, perform coined word verification on the extracted coined word candidates in step S923, refine a coined word extraction pattern and extend a stop word dictionary to improve performance according to whether the coined word candidates are nouns or not as a result of the coined word verification in step S924, and add a coined word, generated through the coined word verification result, to the morpheme dictionary to improve morpheme analysis performance in step S925.
The language analysis processing method based on automatic coined word recognition may correspond to the details described above with reference to
Subsequently, in step S930, keyword extraction may be performed based on a morpheme analysis processing result based on automatic coined word recognition.
In this case, a word corresponding to a noun (a general noun, a proper noun, or the like) in the morpheme analysis processing result may be extracted as a keyword.
Subsequently, in step S940, a candidate may be detected from among advertisement targets registered in a dictionary, based on the extracted keyword.
Subsequently, in step S950, a suitable level and a correlation which match the document content may be calculated by grading a keyword-based priority of the detected candidate advertisement target, and a final advertisement target may be set and recommended.
An embodiment of the present invention may be implemented in a computer system, e.g., as a computer readable medium. As shown in in
Accordingly, an embodiment of the invention may be implemented as a computer implemented method or as a non-transitory computer readable medium with computer executable instructions stored thereon. In an embodiment, when executed by the processor, the computer readable instructions may perform a method according to at least one aspect of the invention.
The language analysis apparatus for automatic coined word recognition and the online contextual advertisement intellectualization apparatus and method based on the same according to the embodiments of the present invention may be implemented in the form of a storage medium that includes computer executable instructions, such as program modules, being executed by a computer. Computer-readable media may be any available media that may be accessed by the computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable media may include computer storage media and communication media. Computer storage media includes both the volatile and non-volatile, removable and non-removable media implemented as any method or technology for storage of information such as computer readable instructions, data structures, program modules, or other data. The medium of communication is a typically computer-readable instruction, and other data in a modulated data signal such as data structures, or program modules, or other transport mechanism and includes any information delivery media.
The method and system according to the embodiment of the present invention have been described in association with a specific embodiment, but their elements, some operations, or all operations may be implemented by using a computer system having general-use hardware architecture.
As described above, according to the embodiments of the present invention, a coined word may be extracted based on a coined word extraction pattern, the extracted coined word may be primarily and automatically verified by using a morphological analyzer, and a morphological dictionary and a coined stop word dictionary may be secondarily and automatically established based on a result of the automatic verification. Accordingly, a morphological dictionary which enables a previously misanalysed morpheme to be accurately analyzed may be provided, and thus, a coined stop word dictionary for decreasing a coined word extraction error may be provided, thereby improving a performance of a coined word extractor.
Moreover, according to the embodiments of the present invention, a morphological analyzer for accurately analyzing coined words included in a dictionary and learning data may be provided to automatically extract coined words, the extracted coined words may be automatically and manually verified, and a performance of each of a coined word extractor and a morphological analyzer is improved by using a result of the verification.
Moreover, according to the embodiments of the present invention, a performance of morpheme analysis technology which is an essential function for recommending an advertisement suitable for content of a document is improved in the advertisement field, and thus, customized advertisements are effectively recommended in the advertisement field where a number of coined words associated with new products appear.
A number of exemplary embodiments have been described above. Nevertheless, it will be understood that various modifications may be made. For example, suitable results may be achieved if the described techniques are performed in a different order and/or if components in a described system, architecture, device, or circuit are combined in a different manner and/or replaced or supplemented by other components or their equivalents. Accordingly, other implementations are within the scope of the following claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 10-2017-0031766 | Mar 2017 | KR | national |
