한국 대중가요 가사 유형화에 토픽 모델링(LDA 및 STM) 적용 실용성 시론 - 70년대 이후 “사랑” 주제어 관련 가사를 중심으로 -
A Study on the Practicability of Categorizing Korean Popular Songs by Topic Modeling(LDA & STM): Based on Lyrics with "Love-centered" Keywords Since 1970s
0
조회
0
다운로드
4
인용
0
참고문헌
키워드
초록
In Korea, popular songs could be a very important way for people to express their "love" feelings, which inspired many researchers to excavate the hidden connections between humans’ love perception and popular songs with different genres based on the lyrics with "love-centered" keywords. Traditionally, a qualitative analysis of "love-centered" lyrics should focus on several typical instances extracted from the whole samples, generalizing the characteristics of a specific writer or brief period. However, with the development of computer science and statistics, it now provides a new perspective for us to elucidate the true meanings behind the natural language texts without human beings’ subjective interpretations. This quantitative analysis method, on the one hand, can efficiently relieve the manual classification burdens and generate a more precise sorting result even than human beings' doing. On the other hand, the rigidly quantized classification criteria sometimes can also incur several enigmatic conclusions conflicting with people's intuitional judgment. To demonstrate the practicability of it, this paper hence chose to manipulate 5082 Korean "love-centered" lyrics collected from online with 2 kinds of considerably valuable "topic models": LAD & STM, aimed to 1. verify the feasibility of automatically constructing a taxonomy system of "love-centered" lyrics; 2. examine and speculate a diachronic correlation between "love-centered" lyrics and published years. As a result, it is of great necessity to optimize the themes K selection when conducting a topic modeling experiment. And based on a reasonable classification, it is possible to observe a significant regressive trend on different groups of "love-centered" lyrics through years. But because of the performance limitation of topic modelling, there are still some biased cases that can be found, which implies a further improvement not just on technical parameter testing but also on a potential collaboration between quantitative and qualitative analysis.
Abstract
In Korea, popular songs could be a very important way for people to express their "love" feelings, which inspired many researchers to excavate the hidden connections between humans’ love perception and popular songs with different genres based on the lyrics with "love-centered" keywords. Traditionally, a qualitative analysis of "love-centered" lyrics should focus on several typical instances extracted from the whole samples, generalizing the characteristics of a specific writer or brief period. However, with the development of computer science and statistics, it now provides a new perspective for us to elucidate the true meanings behind the natural language texts without human beings’ subjective interpretations. This quantitative analysis method, on the one hand, can efficiently relieve the manual classification burdens and generate a more precise sorting result even than human beings' doing. On the other hand, the rigidly quantized classification criteria sometimes can also incur several enigmatic conclusions conflicting with people's intuitional judgment. To demonstrate the practicability of it, this paper hence chose to manipulate 5082 Korean "love-centered" lyrics collected from online with 2 kinds of considerably valuable "topic models": LAD & STM, aimed to 1. verify the feasibility of automatically constructing a taxonomy system of "love-centered" lyrics; 2. examine and speculate a diachronic correlation between "love-centered" lyrics and published years. As a result, it is of great necessity to optimize the themes K selection when conducting a topic modeling experiment. And based on a reasonable classification, it is possible to observe a significant regressive trend on different groups of "love-centered" lyrics through years. But because of the performance limitation of topic modelling, there are still some biased cases that can be found, which implies a further improvement not just on technical parameter testing but also on a potential collaboration between quantitative and qualitative analysis.
참고문헌 (33)
- 강희숙, "K - pop 노랫말의 코드 전환 양상 및 담화 기능", 호남학, (60), pp.193~226 (2016), 호남학연구원
- 곽은희, "감성으로 기억하는 만주 －만주 소재 대중가요 가사를 중심으로－", 만주연구, (18), pp.135~162 (2014), 만주학회[DOI]
- 권지연, "가요에서의 ‘경쾌한’과 ‘감정적’ 감성의 가사 비교 연구", 한국콘텐츠학회 종합학술대회 논문집, 2014(11), pp.149~150 (2014)
- 권지연, "한국대중가요에서의 기쁨과 슬픈 감성을 위한 가사 비교 연구", 한국콘텐츠학회 논문지, 15(3), pp.68~77 (2015), 한국콘텐츠학회[DOI]
- 김광해, "일제 강점기의 대중가요에 대한 계량언어학적 연구 : 유성기 음반 채록본을 중심으로", 한국어 의미, 19(3), pp.197~215 (1998)
- 김익두, "한국 대중가요 연구사 검토", 공연문화연구, (24), pp.5~45 (2012), 한국공연문화학회[DOI]
- 박정은, "현행 교육과정에 의한 『고등학교 음악』의 K-Pop 분석비교연구", 예술교육연구, 12(3), pp.105~123 (2014), 한국예술교육학회
- 배샛별, "일제 강점기 사랑과 이별 소재 대중가요 가사 속 여성 지칭 명사의 특징", 대중서사연구, 22(3), pp.279~306 (2016), 대중서사학회[DOI]
- 배샛별, "일반5: 1945-1959년 한국 대중가요 가사의 특징: 사랑과 이별의 정서를 중심으로", 음악과 현실, (51), pp.183~210 (2016), 민족음악학회
- 서영희, "조명암의 대중가요 가사에 나타난 고향의식", 한민족어문학(구 영남어문학), (64), pp.401~430 (2013), 한민족어문학회
- 손지영, "TextRank 알고리즘을 이용한 음악 가사 요약 기법", 멀티미디어학회논문지, 21(1), pp.45~50 (2018), 한국멀티미디어학회[DOI]
- 안의정, "계량적 접근에 의한 1930-40년대 대중가요 어휘 연구", 어문론총, (63), pp.41~62 (2015), 한국문학언어학회[DOI]
- 윤호진, "한류 20년, 대한민국 빅 콘텐츠" (2016), CommunicationBooks
- 이상욱, "K-pop의 노랫말 특성 연구", 한국문학과 예술, (16), pp.109~142 (2015), 사단법인 한국문학과예술연구소
- 이재환, "가사의 감정 분석과 구조 분석을 이용한 노래 간 유사도 측정", 정보과학회 컴퓨팅의 실제 논문지, 22(10), pp.479~487 (2016), 한국정보과학회
- 장유정, "1970-80년대 한국 대중가요 가사의 특징 -공중파 방송 인기곡을 중심으로-", 공연문화연구, (24), pp.79~113 (2012), 한국공연문화학회[DOI]
- 장유정, "유도순의 대중가요 가사 분석과 작품 규정 문제", 동악어문학, (60), pp.243~271 (2013), 동악어문학회
- Aggarwal, C. C., "Mining text data" (2012), Springer Science &Business Media
- Blei, D. M., "Topic models", Text mining: classification, clustering, and applications, 10(71), pp.34~ (2009)
- Blei, D. M., "Latent dirichlet allocation", Journal of machine Learning research, 3, pp.993~1022 (2003)
- Bok-rae, K., "Past, present and future of Hallyu(Korean Wave)", American International Journal of Contemporary Research, 5(5), pp.154~160 (2015)
- Buntine, W., "Asian Conference on Machine Learning", pp.51~64 (2009), Springer
- Canicatti, A., "Song Genre Classification via Lyric Text Mining", The Steering Committee of The World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp), pp.44~ (2016)
- Feldman, R., "The text mining handbook: advanced approaches in analyzing unstructured data" (2007), Cambridge university press
- Hornik, K., "topicmodels: An R package for fitting topic models", Journal of statistical software, 40(13), pp.1~30 (2011)
- Hotho, A., "A brief survey of text mining", Ldv Forum, 20(1), pp.19~62 (2005)
- Mimno, D., "Optimizing semantic coherence in topic models", Proceedings of the conference on empirical methods in natural language processing, pp.262~272 (2011)
- "ldatuning: Tuning of the latent dirichlet allocation models parameters. R package version 0.2-0" (2016)[URL]
- Roberts, M. E., "Structural topic models for open‐ended survey responses", American Journal of Political Science, 58(4), pp.1064~1082 (2014)
- Wang, H., "Structural topic model for latent topical structure analysis", Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pp.1526~1535 (2011)
- Xia, Xiaoling, "Research on the Model of Lyric Emotion Algorithm", Journal of Physics: Conference Series, 1213, pp.42004~ (2019)
- Yao, L., "Efficient methods for topic model inference on streaming document collections", Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, pp.937~946 (2009)
- Yoo, Y., "Quantitative analysis of a half-century of K‐Pop songs: Association rule analysis of lyrics and social network analysis of singers and composers", Journal of Popular Music Studies, 29(3), pp.e12225~ (2017)