단문의 주제 분석을 위한 LDA와 BTM 토픽모형 평가

문길성(Gil-Seong Mun) 한국자료분석학회 2021 Journal of the Korean Data Analysis Society Vol.23 No.3

토픽 모형은 대량의 문서 집합에서 잠재된 주제를 발견하기 위한 비지도학습의 하나이며, 가장 많이 이용되고 있는 모형은 LDA(Latent Dirichlet Allocation)인 것으로 알려져 있다. LDA는 문서 수준에서 단어의 동시 출현 패턴에 기반을 둔 모형이어서 길이가 긴 문서에서는 효과적으로 이용될 수 있으나, 트윗이나 인스턴트 메시지와 같은 길이가 짧은 문서에는 단어의 동시 출현이 희박하고 자주 사용되지 않는 단어의 출현으로 인하여 양질의 주제 추론이 어렵다는 한계가 있다. 이 문제를 개선하기 위한 하나의 대안으로 BTM(Bi-term Topic Model)이 있다. BTM은 bi-term을 사용하여 단어의 동시 발생을 모델링하고, 주제를 찾기 위하여 이러한 단어의 동시 발생 패턴을 bi-term의 말뭉치에서 집계한다. 본 연구의 목적은 LDA와 BTM의 성능을 비교하고 한국어 기반의 텍스트 데이터에서 BTM의 활용 가능성을 검토하는 데 있다. 이를 위하여 토픽의 질을 양적으로 측정할 수 있는 측도들을 검토하고 실제 자료에 적용하여 두 모형을 평가하였다. 평가결과, 주제 일관성과 문서 분류의 정확성 모두 BTM이 LDA 보다 우수한 것으로 나타났다. 이러한 결과는 단문에서 양질의 토픽을 추출하는 방법으로 BTM의 활용 가능성이 있음을 시사한다. Topic modeling is an unsupervised technique for discovering the underlying topics that occur in a collection of documents, and one popular model is known as LDA (Latent Dirichlet Allocation). It can be used effectively on a variety of long document types because the model implicitly captures the document-level word co-occurrence patterns to reveal topics. The model, however, may have difficulty extracting topics due to the severe data sparsity in short texts such as tweets and instant messages. An alternative to improve this problem is to BTM (Bi-term Topic Model). BTM models the word co-occurrence by using the explicit bi-term, and aggregate these word co-occurrence patterns in the corpus for topic discovering. This study is to compare the performance of LDA and BTM and explore the availability of BTM in Korean-based text data. We reviewed the measures to quantitatively represent the quality of the topics, and evaluated the two models based on real data. As a result, BTM outperformed LDA on short texts, using topic coherence and document classification accuracy as performance measures. The results suggest the availability of BTM to extract topics from short texts.

주제 추출을 위한 맵리듀스 기반의 사전확률 최적화 알고리즘

오선영(SeonYeong Oh),온병원(Byung-Won On) Korean Institute of Information Scientists and Eng 2018 정보과학회논문지 Vol.45 No.5

Various topic extraction algorithms have been used to obtain meaningful information from a large number of text documents. Since the topic extraction algorithms work based on the Bayesian probability model, the prior probabilities, α and β, should be given as inputs. Until now, in order to run the topic extraction models, users have to either take advantage of default prior probability values or determine them subjectively. In this study, we propose a MapReduce-based prior probability optimization algorithm that systematically determines the prior probability values in addition to the improvement of performance and accuracy against a large-scale input data. Unlike the previous single thread algorithm, the proposed MapReduce-based algorithm quickly determines the prior probability values that are suitable for the input data. It then extracts topics with high accuracy after the topic extraction algorithm is executed with the chosen prior probability values. Our experimental results showed that the proposed method outperforms the previous method in the aspect of topic coherence and performance.

한국어 학습자 텍스트의 내용적 오류 양상 - 응집성(coherence)을 중심으로 -

김은정,김혜진 한국작문학회 2023 작문연구 Vol.- No.57

본 연구는 한국어 학습자 쓰기 텍스트에 나타난 내용적 오류를 응집성을 중심으로 객관적으로 진단하여 그 양상을 분석하고 가시화하는 것을 목적으로 한다. 한국어 학습자 텍스트의 경우, 글의 구조가 선명하게 드러나지 않고 필자의 의도를 파악하기 어려울 뿐만 아니라 연구자의 자의적인 해석이 가감되기 때문에 내용 오류를 객관적으로 분석하고 판단하는 것이 쉽지 않다. 이로 인해 한국어 쓰기 교육현장에서 교사들은 내용 오류에 대한 피드백에 어려움을 갖는다. 이러한 쓰기 교육에서의 어려움을 해결해 보고자 본고에서는 텍스트성의 응집성을 중심으로 내용 오류를 관련성 오류, 일관성 오류라고 규명하고, 이러한 오류를 객관적으로 판단하기 위해서 주제부-설명부(TR) 분석과 주제부 전개 유형(TP) 분석을 도구로 사용하여 실제 한국어 학습자의 작문 텍스트를 분석하였다. 이 분석 결과에 따라서 관련성과 일관성의 오류 양상을 유형화할 수 있다. 또한 각각의 오류 양상을 가시화하여 자의적인 해석이 아님을 증명할 수 있었다. 이로써 본 연구는 학습자들의 텍스트에 나타난 응집성의 오류를 가시적으로 드러낸다는 점에서 의의를 가지며, 본 연구를 통해서 교사는 객관적이고 명확하게 설명할 수 있고, 학습자는 쉽게 이해할 수 있는 피드백 방안 마련의 실마리를 제공한 점에서 기여한 바가 있다고 본다.

ESL Writers’ Topical Depths in Persuasive Essays

Yun Bum (Mill) Kim, Hyun Gu Lee 한국영어교과교육학회 2013 영어교과교육 Vol.12 No.2

The purpose of this study is to analyze L2 learners’ writing with diverse cultural backgrounds studying in ESL classes and to identify different aspects in using topical depths of the topical subjects. A total of 53 Asian adult L2 students who applied for the entrance exam for a theological seminary in the Philippines participated in this research. They consisted of 17 Myanmars, 7 Chinese, 18 Koreans, and 11 Filipinos. This study assumed that there would be some different aspects in using topical depths of the topical subjects, as one of the factors that make the writing to be coherent, among the nationality groups and between the lower and higher scored groups. The findings of this study are as follows. First, there was no difference in the percentage of topical depths relatively computed by the number of clauses in any compared groups. Secondly, in correlation analysis, no significant relation was taken noted in relation to the topical depths or the percentage of the topical depths over the number of the clauses. Thirdly, the number of words affects positively on the number of clauses, the number of topical depths, and the scores. It means that sufficient number of words may cause the healthy number of topical depths, which may be positively reflected in the scores.

ESL writers’ topical depths in persuasive essays: Cross cultural observation in relation to the rated scores

김연범,이현구 한국영어교과교육학회 2013 영어교과교육 Vol.12 No.2

The purpose of this study is to analyze L2 learners’ writing with diverse cultural backgrounds studying in ESL classes and to identify different aspects in using topical depths of the topical subjects. A total of 53 Asian adult L2students who applied for the entrance exam for a theological seminary in the Philippines participated in this research. They consisted of 17 Myanmars, 7Chinese, 18 Koreans, and 11 Filipinos. This study assumed that there would be some different aspects in using topical depths of the topical subjects, as one of the factors that make the writing to be coherent, among the nationality groups and between the lower and higher scored groups. The findings of this study are as follows. First, there was no difference in the percentage of topical depths relatively computed by the number of clauses in any compared groups. Secondly, in correlation analysis, no significant relation was taken noted in relation to the topical depths or the percentage of the topical depths over the number of the clauses. Thirdly, the number of words affects positively on the number of clauses, the number of topical depths, and the scores. It means that sufficient number of words may cause the healthy number of topical depths, which may be positively reflected in the scores.

Matthew Arnold's Neo-Platonic Disposition and His Search for the Wholeness

김신희 국제언어문학회 2013 國際言語文學 Vol.- No.28

Matthew Arnold(1822-1888) was a Victorian poet who faced a dilemma between his personal feelings and his assumed social role as a poet. He found that human beings were torn apart and the whole age was divided into multiplicity. As a poet, he felt it his responsibility to give spiritual food to his contemporaries, but found himself not able to do so. Faced with that dilemma, Arnold stopped writing poetry and became a literary and social critic. This paper will show that despite the fact that Arnold devoted his literary career to each different genre, there can be found some consistencies in theme and tone in both his poetry and prose. The consistencies reflect Arnold's Neo-Platonic disposition and it shows that he was devoted to the search for the wholeness of human beings throughout his literary career.

Matthew Arnold"s Neo-Platonic Disposition and His Search for the Wholeness

Kim, Shinhee(김신희) 國際言語文學會 2013 國際言語文學 Vol.- No.28