KR101712507B1 - Smart delivery system and method using wearable device - Google Patents

Abstract

The present invention relates to an entity name linking system of a microblog. According to an embodiment of the present technology, the present invention has an advantage of increasing the accuracy of entity name linking work within a short-length microblog document by performing entity name linking work in consideration of the interest field of a microblog user, real-time social issue, and contextual information of the microblog by using news article or the like, which is a knowledge base and an external resource other than the knowledge base.

Description

Object linking system of microblog. {SMART DELIVERY SYSTEM AND METHOD USING WEARABLE DEVICE}

The present invention relates to an object name linking system for a microblog, and more particularly, to an object name linking system for a microblog, which can perform object name linking operations with higher accuracy, .

Recently, the amount of information is rapidly increasing due to the development of internet and computing technology, the evolution of mobile devices and sensors, and the emergence of networks. Therefore, various studies are being conducted to find the necessary information among the increasing information. Identification of the entity name, which is an area of information extraction, and linking of the recognized entity name to a specific entity are actively attempted to extract meaningful knowledge in vast amount of information. Entity Linking is the task of mapping entity names in text to specific entries in the knowledge base, such as Wikipedia.

Figure 1 shows an example of object linking using a knowledge base. In Fig. 1, the left text box is the object name "Isoara" having bold characters, and the solid arrow "Isoara (singer)" is the correct object name of the knowledge base to be linked.

In order to resolve the ambiguity, the target entity name adjacent context information and the knowledge base are used as external resources. The form of the entity name adjacent context information can be an entry of the knowledge base. For example, in the text box on the left, "Proposal of Lee Soo-Ra" is an entity that exists in the entry of the knowledge base, and it provides the information that the Lee So-young (singer) has performed the program in the knowledge base.

As mentioned above, recent studies on object linking have focused on solving the ambiguity through the knowledge and semantic relation of the real world using the knowledge base as an external resource.

On the other hand, as an example of a tweet that can be described as a typical microblog service, a user communicates through the act of writing a tweet within 140 characters in real time. The tweet is a space of personal interest, daily life, social Issues in real-time.

Object linking using a knowledge base has a good performance in newspaper articles and blog posts. In micro blogs, however, short text lengths and micro blogs are relatively low because of the nature of subjects that are issues in real time, There is a performance problem. Accordingly, there is a need for a method for enhancing object linking performance in a microblog having the above characteristics by using external resources other than the knowledge base.

1. Korean patents (10-2010-00956912)

1. Kim Young-sik, Ham Young-kyun, Kim Ji-sung, Hwang Do-sam and Choi Ki-sun (2014). Detection of object URIs in Korean text: Partial tagging Identification of independent object names and elimination of ambiguities, Proceedings of the 26th Annual Conference on Korean and Korean Information Processing, 100-106.

The object of the present invention is to solve the above problems and to provide an object name linking function of a microblog that performs an object name link operation in consideration of not only a knowledge base but also a context of interest of a microblog user, The purpose of the system is to provide.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and it is an object of the present invention to provide a micro blog system, A context modeling unit for outputting a result of the deconvolution according to the analysis of the weighted object names of the pertinent object names in the perceived entity name by using the knowledge base; An issue model unit for collecting news articles from the web, generating a link to a specific entry of the knowledge base with respect to the collected news articles, and outputting a result of deconvolution according to analysis; And a linking model unit for performing an entity name link operation by integrally analyzing the result of decentralization output from the context model unit and the issue model unit.

The linking model unit may further include a user model unit for outputting a result of the deconvolution by collecting and analyzing the web record created by the microblog text publisher, And a linking model unit for performing an entity name link operation by integrally analyzing the output result of the deconvolution.

Preferably, the context modeling unit outputs the result of the false elimination by calculating the number of link occurrence count of the weighted entity name for each knowledge base page of each candidate entry of the failed entity name, and the user model unit, Constructs a set of entity names in which the entity name corresponding to the knowledge base entry is obtained by extracting all or part of the publisher's past postings from the web and using the left-most long-match method, uni-gram or bi- gram qualities, Wherein the issue modeling unit collects news articles about a predetermined number of days before and after the publication date of the microblog text, and outputs the text of the knowledge base entry page and the news article Using the cosine similarity between the texts of the pages, Wherein the linking modeling unit outputs a result of the weighted solution by performing issue scoring based on a page linking a specific knowledge base entry with respect to each of the output of the context modeling unit, the user modeling unit, and the issue modeling unit, It is possible to solve the property of the object name according to the result of multiplying the variable, to perform the object name link operation, and to use the Wikipedia as the knowledge base.

Preferably, the context modeling unit includes:

는 위키피디아 엔트리에 존재하는 j번 째 후보 개체,

는 중의성 개체가 나타나는 문서 D에 존재하는 i번 째 비중의성 개체임)의 식에 따라 가장 높은 점수를 가지는 중의성 해소 결과를 출력하는 것을 특징으로 하는 것이고, 상기 유저모델부는,

(Where <D> is the set of all entities appearing in linked document D,

Is the j-th candidate entity present in the Wikipedia entry,

Is the i-th weighted entity present in the document D in which the mood entity is represented), and outputs the result of the mood solution having the highest score according to the expression:

는 중의성이 해소되어야 하는 개체명에 대하여 위키피디아 엔트리에 존재하는 j번 째 후보 개체,

는 중의성 개체가 나타나는 문서에 존재하는 i번 째 비중의성 개체임)의 식에 따라 가장 높은 점수를 가지는 중의성 해소 결과를 출력하는 것을 특징으로 하는 것이고, 상기 이슈분석부는,(

Is the j-th candidate entity in the Wikipedia entry for the entity name for which the ambiguity should be resolved,

Is the i-th weighted entity existing in the document in which the mood entity is present), and outputs the result of the mood solution having the highest score according to the equation of (i)

는 위키피디아 엔트리에 존재하는 j번 째 후보 개체,

는

를 주제로 하는 i번 째 뉴스기사,

는 모든

의 집합,

는 중의성 개체가 나타나는 문서 D에 존재하는 k번 째 비중의성 개체임)의 식에 따라 가장 높은 점수를 가지는 중의성 해소 결과를 출력하는 것을 특징으로 하는 것이고, 상기 통합분석부는,(Where <D> is the set of all entities appearing in linked document D,

Is the j-th candidate entity present in the Wikipedia entry,

The

The i-th news article on the theme,

All

A set of,

Is a k-th weighted entity existing in the document D in which the mood entity appears, and outputs the result of the mood solution having the highest score in accordance with the expression:

(E is the name of the object whose ambiguity should be resolved), and can perform the object name link operation.

As described above, according to the present invention, considering the interest field of the microblog user and the contextual information of the microblog and the real-time social issue using the knowledge base and the news source, which are external resources other than the knowledge base, It is possible to increase the accuracy of the object name linking operation in the short-length microblog document.

Figure 1 shows an example of object linking using a knowledge base.
FIG. 2 is a conceptual diagram illustrating a structure of a microblog object name linking system according to an embodiment of the present invention.
3 shows a block diagram of an object name linking system of a microblog according to an embodiment of the present invention.
4 shows a conceptual diagram of a structure of a user model according to an embodiment of the present invention.
FIG. 5 shows a conceptual diagram of the structure of an issue model according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention, and the manner of achieving them, will now be described, by way of example only, with reference to the accompanying drawings, in which: References to examples will be clear. It should be noted that the detailed description of known functions and constructions related to the present invention will not be described in detail when it is determined that the gist of the present invention may be unnecessarily blurred.

In the present invention, an entry can be used in the same sense as an object, and the text of each page represents an object to be described. Also, pages of other entities that have a semantic relationship appear in the text of the page as links. In other words, if there is a link between Wikipedia entities, entities have a semantic relationship. Here, a page can be understood as a web page. Also, in the present invention, object linking is a task of recognizing and linking object names. In the present invention, the knowledge base is preferably a Wikipedia, and even a technology applied to Wikipedia can be applied to other similar knowledge bases. Likewise, even if a tweet is applied to other microblog services similar to tweets, It is possible.

The system proposed in the present invention applies a method of taking into consideration the interests of a microblog user, real-time social issues, and contextual information in tweet remarks.

2 is a conceptual diagram of a structure of a microblog object name linking system according to an embodiment of the present invention.

The proposed system can be composed of three models with reference to FIG. It consists of a Contextual model, a User model, and an Issue model, and the Linking model integrates them.

The context model can resolve the ambiguity by using the objects in the context of the context of the object name with the ambiguity.

The user model can record the microblog account of the user who posted the tweet including the name of the entity that wants to solve the ambiguity, and collect the whole or part of the existing record on the web to eliminate the ambiguity.

Issue models can extract qualities not covered by a knowledge base such as Wikipedia from news articles that are external resources.

The linking model analyzes the output of all three models, determines the most likely candidate entry, and links the object name.

The present invention may include a context model unit, a user model unit, an issue model unit, and a linking model unit, each of which includes an object name that recognizes an entity name and the above-described respective models.

3 shows a block diagram of an object name linking system of a microblog according to an embodiment of the present invention.

3, the object name linking apparatus of a microblog according to an exemplary embodiment of the present invention includes an expression unit 100, a context model unit 200, an issue model unit 300, a linking model unit 500 ). In addition, it may further include a user model unit 400.

The expression unit 100, which is an object name, collects the microblog text and recognizes the entity name. For example, the food unit 100, which is an object name, collects microblog texts for performing an object name linking operation from the web and recognizes object names included in the collected microblog texts.

For object linking, object identification is performed because it is necessary to identify the objects to be targeted. Entity-name recognition in knowledge base-based object linking may be to extract candidate entries of normalized names of entities in all possible knowledge bases in the text. For example, in the object name recognition of the object named 'Isoara', it is classified into the class such as' PER 'in the recognition of the object name, but in the recognition of the knowledge base object name, the object name such as' Extract the entry.

Here, it is possible to extract only those corresponding to the object name dictionary among all possible n-gram terms in the text document, or to learn the object name by learning the learning document to which the object name is correctly assigned as the classifier.

In Korean language linking, it is possible to construct a surface form in which each object name can appear in advance, and use the object boundary recognition method using SVM. The recognition of the entity name in the tweet can be attempted in a ring way by using the KNN classifier and the CRF labeler in a hybrid manner for an English tweet.

After recognizing the entity name, it is necessary to resolve the identity of the entity name having the correctness among the recognized entity names for accurate object linking.

는 위키피디아 엔트리에 존재하는 j번 째 비중의성 개체,

는 E가 링크 될 수 있는 위키피디아 엔트리에 존재하는 j번 째 후보 개체를 의미한다.In the equation of the present invention, E is the name of the object for which the ambiguity is to be resolved,

Is the j-th weighted object in the Wikipedia entry,

Is the j-th candidate entity in the Wikipedia entry where E can be linked.

는

를 주제로 하는 i번 째 뉴스기사,

는 모든

의 집합, [e]는 위키피디아 엔트리 e가 가지는 위키피디아 링크의 집합,

는 각 j번 째 후보 엔트리의 문맥 모델, 유저 모델, 이슈모델의 스코어를 의미한다. Also, <D> is the set of all objects appearing in document D,

The

The i-th news article on the theme,

All

, [E] is the set of Wikipedia links that Wikipedia entry e has,

Is the score of the context model, the user model, and the issue model of each jth candidate entry.

The context modeling unit 200 outputs the result of the deconvolution by analyzing the weighted object names of the surrounding contexts of the object names among the object names recognized by the expression unit 100, which is the object name, using the knowledge base.

The context modeling unit 200 uses context information. Here contextual information refers to contiguous object names of contiguous individuals of the same name. For example, if there is an entity name in a tweet that has a neutrality, all the non-weighted entity names included in the tweet are relevant.

The semantic associations used in conventional technologies using most Wikipedia as external resources use Wikipedia category information as well as adjacent term groups. However, when using Wikipedia category information, performance may be lower than when it is not used.

The user of the Wikipedia or the Korean Wikipedia may not use the category information in the embodiment of the present invention because the category information is insufficient relative to the English Wikipedia.

The context modeling unit 200 may use a scoring method as shown in Equation 1 below.

Equation 1

로 나타낼 때 문맥 모델부(200)는

가 후보 엔트리들 중에서 가장 높은 점수를 가지는 엔트리를 중의성 해소 결과로 출력할 수 있다.The name of the object having the affinity and the degree of correlation

The context modeling unit 200 determines

Can output the entry having the highest score among the candidate entries as a result of the vanity elimination.

은

과

의 가중 합으로 이루어진다.

은 중의성 개체가 나타나는 문서에 존재하는 i번 째 비중의성 개체

의 위키피디아 페이지에 나타나는 링크와 후보 개체

의 위키피디아 페이지의 링크 정보가 겹치는 횟수를 계산하는 수식이다.

silver

and

.

Is the i-th weighted object

Links appearing on the Wikipedia page of

This is a formula that calculates the number of times the link information of the Wikipedia page is overlapped.

가 각각 “이소라(가수)”, “이소라(모델)”, “이소라(배구선수)”에 대해 계산되고, 이 때,

는 중의성 개체가 발견된 트윗에서 나타난 비중의성 개체가 된다. 문맥모델부(200)는

가 가장 높은 점수를 가지는 후보 엔트리로 링킹이 가능하도록 중의성 해소 결과로서 출력을 할 수 있으며, 엔트리와 함께 그 점수를 포함하여 출력할 수 있다.Wikipedia has the same name "disambiguation page". Referring to FIG. 0, the document having the same name as &quot; Isola &quot; has three candidates. The first "Isola" is the model of the Republic of Korea, and the second is the singer of the Republic of Korea. The last "Isola" is a Korean volleyball player. For example, if E is "isola," then E

Are computed for "Isola (singer)", "Isola (model)" and "Isola (volleyball player)", respectively,

Becomes a weighted object in the tweet where the mural object is found. The context modeling unit 200

Can be output as a result of the false elimination so that linking to the candidate entry having the highest score can be performed, and the output including the score can be output together with the entry.

In addition, the apparatus for linking object names of microblogs according to the present invention may include a user model unit 300.

The user modeling unit 300 outputs the result of the deconvolution according to the collection and analysis of the web record created from the microblog text publisher. Web history written by a microblog text publisher includes records written by the publisher on the microblog as well as other various webs.

Micro blogs, such as tweets, are characterized by the ability to communicate about their interests. Using these characteristics, the user model can use all or a portion of the user's previous tweets and other web records that can be collected to deal with the user's behavior and interests. It is assumed that the user has interest and interest in the object names appearing in the user's previous tweet or the like.

FIG. 4 shows a conceptual diagram of the structure of the user model unit 300 according to the present invention.

Referring to FIG. 4, for example, if the user model unit 300 recognizes an object name having a falsehood in a specific tweet, all tweets of the user who left the tweet are extracted. The Entity Intergration module extracts and adds to the <D> if there is a corresponding entry in the Wikipedia entry dictionary built using the left-most matching method and the uni-gram and bi-gram features. D's Wikipedia entries are mostly made up of noun phrases. However, due to the nature of the tweet data, the performance of the morphological analysis deteriorates due to the large number of noises. Therefore, uni-grams and bi-grams can be used as the qualities.

The user modeling unit 300 can calculate the score of the user model using the following equation (2).

Equation 2

를 추출한다. 이 자질들은 링크 후보 개체인

중 ”이소라 (가수)“가 최종적으로 중의성 해소 결과로 출력될 가능성을 높여준다. 왜냐하면 ”이소라 (가수)“의 위키피디아 페이지에는 ”나가수“와 ”MBC“라는 개체명을 링크로 가지고 있기 때문에

(이소라 (가수))의 점수가 높아지기 때문이다.In accordance with the example of FIG. 4, an entity having a meaningfulness, &quot; isola, &quot; is found in the tweet of the user p2 and collects all past tweets of the p2. All past tweets collected at this time can be viewed as D. The object integration module is an object name such as "Nagasu" in which Iora (singer) appeared and "MBC"

. These qualities are linked candidate objects

(Singer) "is finally output as a result of resolution of ambiguity. Because the Wikipedia page for "Isola (singer)" has a link to the object name "Nagasu" and "MBC"

(Lee So-ra (singer)) score is higher.

The issue modeling unit 500 collects news articles from the web, generates a link to a specific entry of the knowledge base for the collected news articles,

The issue model unit 500 can handle contextual models and real-time social events that the user model can not handle, or events that are of interest to the trivial public. For example, events that happen in real time by famous entertainers, or very minor events, are not covered by a heavy knowledge base like Wikipedia, but they are topics that are covered in many blogs such as tweets. Since the news article has the characteristics of dealing with the above-mentioned events, the issue model unit 500 uses the news article as an external resource.

5 is a conceptual diagram illustrating a structure of an issue model implemented by a news linking module and an entity name generation module according to an embodiment of the present invention.

Referring to FIG. 5, when the issue model recognizes the object name "Isora" having the attribute of a substance in the same manner as the issue model, "IsoRa" is recorded in k days of the tweet posting date on which the news linking module is recognized Collect the news articles included in the title.

와 링킹 된다. The collected news articles have texts similar to the text displayed on the pages of Wikipedia entries "Isola (singer)", "Isola (model)" and "Isola (volleyball player) Lt; RTI ID = 0.0 &gt;

Lt; / RTI &gt;

,

,

가 각각 "이소라 (가수)”，"이소라 (모델)” 그리고 "이소라 (배구선수)”가 된다. 다시 말해 이슈모델은 뉴스 기사 각각 특정 위키피디아 엔트리와 링크함으로써 뉴스 기사 하나를 하나의 위키피디아 페이지로 취급한다. 각 기사는 후보 엔트리

를 주제로 하는 i번 째 뉴스 기사

가 된다. 개체 생성 모듈(Entity Generative module)은 각 뉴스 기사마다 위키피디아 링크를 생성한다. 신문 기사 속 작은 따옴표는 이름 표시 기능을 가질 수 있고, 이름 표시를 할 때 신문사마다 약간의 차이가 있으나, 책 이름, 영화 이름, 음반 이름, 드라마 이름 등이 작은 따옴표로 묶일 수 있다. 개체 생성 모듈은 뉴스 기사의 작은 따옴표로 명시된 용어들을 의미 있는 개체명으로 가정하고, 모두 링크로 생성할 수 있다. In the above example,

,

,

The issue model will link each news article to a specific Wikipedia entry so that one news article is treated as a single Wikipedia page (eg, Each article has a candidate entry

I-th news article on

. The Entity Generative module creates a Wikipedia link for each news article. Single quotes in a newspaper article can have a name display function, and there are slight differences between newspapers when displaying names, but book names, movie names, album names, and drama names can be enclosed in single quotes. The entity creation module can generate all links, assuming that the terms specified by the single quotation marks of the news article are meaningful entity names.

Table 1 below shows an example of an entity that can be generated according to the entity creation rule and the example of FIG.

Table 1

인 “이소라 (모델)”로 링킹 될 수 있는 중요한 개체 정보가 될 수 있다. 두 번째 생성 규칙은 뉴스 기사를 형태소 분석하여 명사 자질만 사용하여, 명사 uni-gram, bi-gram이 위키피디아 개체명 사전에 존재할 경우 추가한다. 예를 들면, “컬투”, “정찬우”, “모델”이 해당된다. 최종적으로 이슈모델부()는 아래의 수학식 3과 같이 후보 엔트리들을 스코어링 한다. 표 1의 예에 따라 적용한다면,

가 “이소라 (모델)”일 때,

는 “이소라 (모델)”을 제목에 포함하는 i번째 뉴스 기사이고, 개체 생성 모듈의 링크 생성 규칙에 의해 뉴스 기사에 링크를 생성하여 트윗을 남긴 사용자의 기록을 모은 트윗 문서 D에 나타나는 비중의성 개체들이 위키피디아 페이지에 나타나는 링크 정보의 동시 출현 횟수를 점수에 반영한다.As shown in Table 1, the "Salvation Table" is actually an entry that does not exist in Wikipedia, but the title of the program that the "

Can be important entity information that can be linked to &quot; isola (model) &quot; The second generation rule uses the noun qualities only, stemming from news articles, and adds nouns uni-grams and bi-grams if they exist in the Wikipedia object name dictionary. For example, "Kultu", "Jeong Chan-woo", "Model" are applicable. Finally, the issue model unit scans the candidate entries as shown in Equation (3) below. If applied according to the example of Table 1,

Is "Isola (model)",

Is the i-th news news item that includes "Isoara (model)" in the title, and is a tweet article in the tweet document D that collects the record of the user who left the tweet by generating a link to the news article by the link generation rule of the object creation module Reflects the number of simultaneous occurrences of link information appearing on the Wikipedia page to the score.

Equation 3

The linking modeling unit 500 performs an entity name linking operation by performing integrated analysis on the result of decentralization output from the context modeling unit 200 and the issue modeling unit 400. [ In addition, for the more accurate object name linking, the result of the decentralized solution output from the user modeling unit 300 is further integrated and analyzed together with the results of the context modeling unit 200 and the issue modeling unit 400, You can also do the work.

,

,

를 이용하여 계산할 수도 있다.For example, the linking model unit 500 may integrate the points included in the result of deconvolution outputted from the context model unit 200, the user model unit 300, and the issue model unit 400, It may be to calculate the final score. The final score is calculated using the weight parameter &lt; RTI ID = 0.0 &gt;

,

,

. &Lt; / RTI &gt;

Equation 4

The weight parameters are the degree to which Twitter users reflect the interest of the user considering the context model, user model and issue model, and the degree to reflect the real-time issue in the tweet, Can be determined.

Experiment and evaluation

In the embodiment of the present invention, the experiment object is a person entity name having a meaning in a document having the same name as Wikipedia, and the object name dictionary having the same name is constructed using Wikipedia category information for the experiment.

,

,

값은 실험 데이터에 의존적인 매개 변수로 실험적으로 결정하였다.(

=0.4,

=0.35,

=0.25)In order to construct experimental data, we collected 1,6367 tweets totaling 5060 tweets per 300 users who actively use tweets. Among the collected tweets, tweets were selected that include the object names in the object name dictionary having the same name. In order to verify the reliability, three tweets were used to tweet tweets with the same name. Twenty - four Korean tweets were constructed. A total of 33 names with the same name appear in the 248 identical names in the collected 248 tweet data, and an average of 3.45 people in the collected tweet data with 33 names were observed. In the Wikipedia document On average, 4.75 people with the same name were observed. The news articles used in the issue model collected the amounts for three days before and after the registration of the tweets containing the mysterious object names (k = 3)

,

,

The values were determined experimentally as parameters dependent on the experimental data. (

= 0.4,

= 0.35,

= 0.25)

The knowledge base for the whole system used Korean Wikipedia and the "Jhannanum" morpheme analyzer was used for morphological analysis of news articles and Wikipedia documents in the issue model. In addition, pre-processing was performed to delete all the tweet data and the unnecessary data of the Wikipedia document before proceeding with the experiment.

In Table 2 below, an accuracy index (number of data determined by the system as the correct answer / total number of data) is used to evaluate the performance of the system according to an embodiment of the present invention, and the performance is compared for each system added . When the proposed models were added based on the context model using only the knowledge base, the performance was significantly increased and the final system performance was 67.7%. In the experimental data, the accuracy is very high considering that the average individuals with the mean of 4.75 indicate the same name.

Table 2

Table 3 below shows the result of measuring whether the issue model of the issue model unit according to an embodiment of the present invention correctly links a news article to an answering Wikipedia page. A total of 836 news articles collected were tagged by two experimen- tal cross-validators and tagged with the correctness of the name of the subject in the news article title, and the accuracy of the experiment was 70.2%.

Table 3

Table 4 below shows an example of increasing the accuracy of false elimination by adding a user model.

Table 4

In the first example in Table 4, it can be seen from the names of objects such as "Samsung", "Lotte", "LG", "Baseball field" extracted from the previous tweet that the user of the tweet usually liked the baseball, The names of the "Samsung" and "Lotte" are linked to the Wikipedia page "Kim Tae-gyun (1971)". Extracted object names from the issue model further increased the final score so that they could be linked to "Kim Tae Kyun (1971)". In the second example of Table 4, unlike the first example, although the object model could not be extracted as a clue in the user model, the "joinability" can be obtained by extracting the object names "drama", "movie" To be linked to "Jo In Sung (actor)".

According to a preferred embodiment of the present invention, a microcomputer including a context modeling unit 200, a user modeling unit 300, an issue modeling unit 400, and a linking modeling unit 500 integrating the context modeling unit 200, A blog object name recognition system can be provided. The proposed system can overcome the limit by using the news articles as external resources for trivial events or real - time events not covered by the knowledge base such as Wikipedia and can consider the user 's interest by using user records.

While the present invention has been particularly shown and described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. It will be appreciated that many changes and modifications may be made without departing from the invention. Accordingly, all such appropriate modifications and changes, and equivalents thereof, should be regarded as within the scope of the present invention.

100: entity name recognition unit
200: Context model part
300: User model part
400: issue model part
500: Linking model part

Claims (4)

Which is an object name that collects micro blog text and recognizes the object name;
A context modeling unit for outputting a result of the deconvolution according to the analysis of the weighted object names of the pertinent object names in the perceived entity name by using the knowledge base;
An issue model unit for collecting news articles from the web, generating a link to a specific entry of the knowledge base with respect to the collected news articles, and outputting a result of deconvolution according to analysis;
A user model unit for outputting a result of decentralization by collecting and analyzing a web record created from a micro blog text publisher; And
And a linking model unit for performing an entity name link operation by integrally analyzing the result of decentralization output from the context model unit, the user model unit, and the issue model unit,
The context modeling unit,
And outputting the result of deconvolution by calculating the number of link occurrence counts of the weighted entity name for each knowledge base page of the candidate entities with the permissible entity name,
The user model unit,
Constructs a set of entity names in which the entity name corresponding to the knowledge base entry is obtained by extracting all or part of the publisher's past postings from the web and using the left-most long-match method, uni-gram or bi- gram qualities, And outputs the result of deconvolution according to the extraction frequency of the non-weighted object name from the post,
The issue model unit,
Collects news articles about a predetermined number of days before the publication date of the microblog text, and generates a page linking a specific knowledge base entry for each news article using the cosine similarity between the text of the knowledge base entry page and the text of the news article page Based on the issue score, it outputs the result of deconvolution,
The linking model unit,
Wherein object name linking is performed by eliminating the identity of the entity name according to a result of multiplying the output of the context model unit and the output of the issue model unit by a weight parameter.
delete delete The method according to claim 1,
The context modeling unit,

(Where <D> is the set of all entities appearing in linked document D,

Is the j-th candidate entity present in the Wikipedia entry,

Is the i-th weighted entity present in the document D where the mismatch entity appears)
And outputs the result of deconvolution having the highest score according to the expression of &quot;
The user model unit,

(

Is the j-th candidate entity in the Wikipedia entry for the entity name for which the ambiguity should be resolved,

Is the i-th weighted entity present in the document in which the mismatch entity appears)
And outputs the result of deconvolution having the highest score according to the expression of &quot;
The issue model unit,

(Where <D> is the set of all entities appearing in linked document D,

Is the j-th candidate entity present in the Wikipedia entry,

The

The i-th news article on the theme,

All

A set of,

Is the k-th weighted entity present in the document D where the ambiguous entity appears)
And outputs the result of deconvolution having the highest score according to the expression of &quot;

The linking model unit,
(E is the name of the object whose ambiguity should be resolved)
And the object name linking operation is carried out in the object name linking system of the micro blog.

Images