인공 지능형 뉴스 제작을 위한 뉴스 언어 데이터베이스 개발 연구 : 일간 신문과 지상파 방송 기사에 대한 통계 학습을 중심으로
초록
로봇 기자가 오탈자와 오정보 등 심각한 오류가 없이 기사를 빠르고 정확하게 제작하면서, 국내외 많은 언론사의 뉴스 편집국에 로봇 기자가 등장했다. 이 상황에서 로봇 기자의 능력을 개선하는 데에 필요한 체계적인 데이터베이스가 필요하다. 본 연구는 이러한 요구에 부응해 로봇 기자의 구성 요소에 도움을 줄 뉴스 언어 데이터베이스를 개발하고자 한다. 로봇 기자는 이 데이터베이스를 장착해 구조, 범주, 변수 등 구체적 개념으로 뉴스 텍스트를 분석할 수 있다. 본 연구는 이러한 데이터베이스를 개발하기 위해 국내 주요 일간 신문 10개 사와 3개 지상파 방송사가 보도한 사건 기사, 기획 기사, 사설을 두 번의 2주간 무작위 설정 주간에 걸쳐 수집해 이를 훈련 자료로 사용했다. 먼저 본 연구는 구조, 범주, 하위 범주, 변수 등 이론적 틀을 선행 연구와 연구 논리 등을 바탕으로 추출했다. 이 결과, 의미 구조, 이야기 구조, 수사적 구조, 구문론적 구조에 인물, 자원, 조직, 지식, 내용, 시제, 술어 형태 등 범주가 출현했고, 범주별로 공인, 이전 정보, 상황/장면 설명, 후속 결과, 구어성, 외래어 등 40개의 변수가 등장했다. 그러나 이 이론적 틀을 훈련 자료에 투입했을 때, 19,607개 단어가 행위, 능동형 등 31개 변수에 속하는 것으로 나타났다. 분석 대상인 언론사 유형을 주요 일간 신문과 지상파 방송사에서 종합편성채널, 보도전문채널, 뉴스 통신사, 인터넷 언론사, 지역 언론사 등으로 확장하면, 본 연구가 제안한 40개 변수에 해당하는 단어가 나올 가능성이 있다. 본 연구는 31개 변수와 대규모 단어로 구성된 뉴스 언어 데이터베이스를 검증하기 위해 <한국경제신문>과 <한국일보>가 코로나19 국내 첫 확진자가 나온 2020년 1월 20일부터 1년간 보도한 사설 98건을 수집했다. <한국경제신문> 사설에는 26개 변수, <한국일보> 사설에는 23개 변수가 각각 등장했는데, 2개 신문의 사설은 문어적이고 중립적이며 이전 정보를 가장 많이 사용했다. 본 연구의 가장 큰 의의는 구조, 범주, 변수, 단어 등을 포함한 뉴스 언어 데이터베이스가 국내 언론학에 드물어 참신하고 독특한 가치가 있으며, 뉴스 언어를 이해할 때 도움이 된다는 점이다. 또한, 이 데이터베이스는 기자들이 뉴스 언어와 기사문을 어떻게 구성하는지를 유용하게 설명할 수 있으며, 본 연구 결과는 인공 지능형 뉴스 제작에 유의하게 이바지할 수 있다. 가령, 로봇 기자는 기사 작성의 논리를 평가하고 훈련용 기사에서 이러한 논리를 학습할 수 있다. 이후에 로봇 기자는 실제 기사를 일관되고 자동으로 작성하는 데에 이 논리를 적용할 수 있다.
Abstract
Given their fast and accurate production with no significant errors (e.g., typos and misinformation), robot journalists have entered many newsrooms and secured their presence in Korea and abroad. This situation asks for a more systematic database for enhancing the ability of robot journalists. Given this need, this study develops a news language database, which can make up the elements of robot journalists. By installing the database, robot journalists can analyze news texts in terms of specific concepts, such as structure, category, and variable. For this purpose, as training data, this study collected massive amounts of straight stories, features, and editorials from 10 major daily newspapers and 3 major television networks via their corresponding websites during two randomly constructed weeks, twice. Before detailed analysis, this study first extracted a theoretical model, such as structures, categories, sub-categories, and variables, based on previous studies and research rationales. As a result, for semantic, script, rhetoric, and syntactic structures, person, resource, group, knowledge, content, tense, and verb form emerged. For each category emerged, public person, previous information, explanation of situation/scene, follow-up, orality, and foreign words. However, these theoretical frameworks applied to the training data, resulting in 31 variables, such as actions and active comprising 19,607 words. If the unit of analysis would include comprehensive programming channels, news channels, news agencies, Internet news media, and local news media, the 40 variables proposed in the theoretical model could emerge. This study collected one-year editorials (98) from Hankyung and Hankook Ilbo since the first COVID-19 patient to test the validity of the database containing 31 variables and the many words. In the Hankyung editorials, 26 variables appeared, while 23 variables emerged from those of the Hankook llbo. The two newspapers used literal words, took neutral tones, and expressed previous information in the editorials most frequently. The major contribution of this study includes that such a news language database is a rare case in the journalism discipline of Korea, which comprises structures, categories, variables, and words. So, the database provides fresh and unique values and insights into news language. Further, the database has practical utilities for explaining how journalists construct news language and news texts. The findings show meaningful contributions to artificial intelligence-based news-making in systematic ways. For instance, robot reporters could evaluate the logical patterns of story-writing and learn them from training news stories. They may apply such logical patterns to writing a story in a consistent and automatic manner.
Keywords:
News Language Database, Artificial Intelligence-Based News Making, Story Structure, Variable, Statistical Learning키워드:
뉴스 언어 데이터베이스, 인공지능 기반 뉴스 제작, 기사 구조, 변수, 통계 학습Acknowledgments
This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea(NRF-2020S1A5A2A01040590) (이 논문은 2020년 대한민국 교육부와 한국연구재단의 인문사회분야 중견연구자지원사업의 지원을 받아 수행된 연구임(NRF-2020S1A5A2A01040590)).
References
- Ahn, E. J. (2008). A study on the orality score of words and its application using corpus. Journal of the Society of Korean Language and Literature, 57, 93-119.
- An, S.-K., & Gower, K. K. (2009). How do the news media frame crises? A content analysis of crisis news coverage. Public Relations Review, 35(2), 107-112. [https://doi.org/10.1016/j.pubrev.2009.01.010]
- Baek, S. G. (2003). Coverage patterns of Korea TV on U.S.-IRAQ war and its meaning structures: With an emphasis of narrative and meaning structure of TV pictures. Studies of Broadcasting Culture, 15(1), 117-158.
- Bell, A. (1998). The discourse structure of news stories. In A. Bell & P. Garrett (Eds.), Approaches to media discourse (pp. 64-104). Malden, MA: Blackwell Publishing.
- Blake, K. (2006). Inverted pyramid story format. https://kelab.tamu.edu/SPB_Encylopedia/data/Inverted%20pyramid%20story%20format.pdf
- Carley, K. M., Diesner, J., Reminga, J., & Tsvetovat, M. (2004). An integrated approach to the collection and analysis of network data. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.61.8253&rep=rep1&type=pdf
- Carlson, M. (2015). The robotic reporter: Automated journalism and the redefinition of labor, compositional forms, and journalistic authority. Digital Journalism, 3(3), 416-431. [https://doi.org/10.1080/21670811.2014.976412]
- Chan, C.-H., Wessler, H., Rinke, E. M., Welbers, K., Van Atteveldt, W., & Althaus, S. (2020). How combining terrorism, muslim, and refugee topics drives emotional tone in online news: A six-country cross-cultural sentiment analysis. International Journal of Communication, 14, 3569-3594. [https://doi.org/10.1080/19312458.2020.1812555]
- Clerwall, C. (2014). Enter the robot journalist: Users’ perceptions of automated content. Journalism Practice, 8(5), 519-531. [https://doi.org/10.1080/17512786.2014.883116]
- Corman, S. R., Kuhn, T., Mcphee, R. D., & Dooley, K. J. (2002). Studying complex discursive systems: Centering resonance analysis of communication. Human Communication Research, 28(2), 157-206. [https://doi.org/10.1111/j.1468-2958.2002.tb00802.x]
- Diesner, J., & Carley, K. M. (2005). Revealing social structure from texts: Meta-matrix text analysis as a novel method for network text analysis. In V. K. Narayanan & D. J. Amstrong (Eds.), Causal mapping for research in information technology (pp. 81-108). IDEA Group Publishing. [https://doi.org/10.4018/978-1-59140-396-8.ch004]
- Dörr, K. N. (2016). Mapping the field of algorithmic journalism. Digital Journalism, 4(6), 700-722. [https://doi.org/10.1080/21670811.2015.1096748]
- Dunn, A. (2003). Telling the story: Narrative and radio news. Radio Journal: International Studies in Broadcast & Audio Media, 1(2), 113-127. [https://doi.org/10.1386/rajo.1.2.113/0]
- Entman, R. M. (2010). Media framing biases and political power: Explaining slant in news of Campaign 2008. Journalism, 11(4), 389-408. [https://doi.org/10.1177/1464884910367587]
- Fowler, R. (2013). Language in the news: Discourse and ideology in the press. Oxon, UK: Routledge. [https://doi.org/10.4324/9781315002057]
- Gamson, W. A. (1992). Talking politics. Cambridge, UK: Cambridge University Press.
- Gans, H. J. (2011). Multiperspectival news revisited: Journalism and representative democracy. Journalism, 12(1), 3-13. [https://doi.org/10.1177/1464884910385289]
- Ha, S.-H., & Lee, M.-K. (2012). A news frame analysis by the South Korean press on the livelihoods of a North Koreans. Korean Journal of Communication & Information, 58, 222-241.
- Hester, J. B., & Dougall, E. (2007). The efficiency of constructed week sampling for content analysis of online news. Journalism & Mass Communication Quarterly, 84(4), 811-824. [https://doi.org/10.1177/107769900708400410]
- Iedema, R. A. M. (1997). The structure of the accident news story. Australian Review of Applied Linguistics, 20(2), 95-118. [https://doi.org/10.1075/aral.20.2.06ied]
- James, G, Witten, D, Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning with applications in R. New York, NY: Springer. [https://doi.org/10.1007/978-1-4614-7138-7]
- Jeon, H. (2016). KoNLP: Korean NLP Package. R package version 0.80.1.https://CRAN.R-project.org/package=KoNLP
- Jung, H. (2011, January 8). Population is similar to that of California, its size is similar to that of New Jersey. JoongAng Ilbo. http://www.koreadaily.com/news/read.asp?art_id=1139586
- Kim, D., & Lee, J. (2015). Robot journalism: Algorithmic approach to automated news article generation. Korean Journal of Journalism & Communication Studies, 59(5), 64-95.
- Kim, E-J., & Bang, J.-B. (2010). Change of news frames in news agencys reports on elections: Focusing on Yonhap News Agencys reports on the presidential elections. Korean Journal of Broadcasting and Telecommunication Studies, 24(5), 90-125.
- Kim, H. (2017). An analysis of reporting constructions in newspaper report texts. Textlinguistics, 42, 1-29.
- Kim, H. S. (2017). Korean natural language processing techniques – Past and present. New Korean Language Life, 27(4), 81-98.
- Kitch, C. (2003). “Mourning in America”: ritual, redemption, and recovery in news narrative after September 11. Journalism Studies, 4(2), 213-224. [https://doi.org/10.1080/1461670032000074801]
- Klinenberg, E. (2005). Convergence: News production in a digital age. The ANNALS of the American Academy of Political and Social Science, 597(1), 48-64. [https://doi.org/10.1177/0002716204270346]
- Ko, Y.-C. (2015). A comparison of the way of composing news contents on the front pages of local newspapers that are being published in South Korea and the United States: Focus on the types of news articles, the forms of news writing, photos and infographic. Journal of Communication Science, 15(1), 5-47. [https://doi.org/10.14696/jcs.2015.03.15.1.5]
- Korea ABC (2019). 2019 paid published circulation of newspapers. http://www.kabc.or.kr/about/notices/100000002887
- Korea Press Foundation (2018). 2018 media audience’s perception analysis: 23th examination of user patterns in changes in the media environment (report 2018-02). Seoul: Korea Press Foundation.
- Korea Press Foundation (2019). Media statistics: Number of employees in the media industry. https://www.kpf.or.kr/front/mediaStats/mediaStatsDetail.do
- Kotenidis, E., & Veglis, A. (2021). Algorithmic journalism – Current applications and future perspectives. Journalism and Media, 2, 244-257. [https://doi.org/10.3390/journalmedia2020014]
- Latar, N. L. (2014). The robot journalist in the age of social physics: The end of human journalism? In G. Einav (ed.), The new world of transitioned media: Digital realignment and industry transformation (pp. 65-80). New York, NY: Springer. [https://doi.org/10.1007/978-3-319-09009-2_6]
- Lee, G., & Park, J. (2006). A frame analysis of news coverage on the kimchie risk in 2005. Korean Journal of Broadcasting and Telecommunication Studies, 20(5), 260-305.
- Lee, H.-H., Lee, J.-K., Choi, J.-H., Cheong, S.-H., & Kang, K.-S. (2015). How does the Korean press see the Korean unification?: Focusing on the content analysis of news coverage about Korean unification by each regime, type of media, and press. Korean Journal of Broadcasting and Telecommunication Studies, 29(2), 220-259.
- Lee, S. (2013). Texttypologische Aspekte der Fernsehnachrichtensendungen. Textlinguistic, 35, 207-228.
- Lee, W.-S., Bae, J.-Y., & Lim, B.-S. (2015). Structural analysis on the news components of the local newspaper: Focusing on the contents and forms of the Busan Ilbo. Journal of Communication Science, 15(3), 221-266. [https://doi.org/10.14696/jcs.2015.09.15.3.221]
- Leppänen, L., Munezero, M., Granroth-Wilding, M., & Toivonen, H. (2017, September 4-7). Data-driven news generation for automated journalism. Proceedings of the 10th International Natural Language Generation Conference (pp. 188-197). [https://doi.org/10.18653/v1/W17-3528]
- Lim, B.-S., & Lee, W.-S. (2011). The influence of newsroom integration upon process of engaging in journalism. Korean Journal of Communication & Information, 53, 29-52.
- Lim, D. (2003). On the Modal System in Modern Korean. Korean Semantics, 12, 127-153.
- Lim, J. (2010). Analysis of the nature of intermedia agenda-setting effects among major news web sites. Journal of Communication Science, 10(4), 498-532.
- Lim, J. (2019). An exploratory study of automatic extraction of news frames in terms of the coverage of television networks, comprehensive programming channels, and news channels of government’s real estate policy. Broadcasting & Communication, 20(1), 47-96.
- Ministry of Health and Welfare (2020). Status of corona virus (COVID-19) domestic cases (Regular briefing on March 27). http://www.mohw.go.kr/react/al/sal0301vw.jsp?PAR_MENU_ID=04&MENU_ID=0403&page=1&CONT_SEQ=353770
- Narrative Science (2016). Data-driven communication at machine scale. https://www.narrativescience.com/quill
- Pan, Z., & Kosicki, G. M. (1993). Framing analysis: An approach to news discourse. Political Communication, 10(1), 55-75. [https://doi.org/10.1080/10584609.1993.9962963]
- Pang, H., & Lee, G. (2016). The way of producing the image of North Korea, exploited by the channels of comprehensive programing in South Korea. Korean Journal of Journalism & Communication Studies, 60(2), 338-365. [https://doi.org/10.20879/kjjcs.2016.60.2.012]
- Riffe, D., Aust, C. F., & Lacy, S. R. (1993). The effectiveness of random, consecutive day and constructed week sampling in newspaper content analysis. Journalism & Mass Communication Quarterly, 70(1), 133-139. [https://doi.org/10.1177/107769909307000115]
- Roberts, M., & McCombs, M. E. (1994). Agenda setting and political advertising: Origins of the news agendas. Political Communication, 11(3), 249-262. [https://doi.org/10.1080/10584609.1994.9963030]
- Sadia, S., & Ghani, M. (2019). Modality in editorials of Pakistani English newspapers: A corpus based study. International Journal of English Linguistics, 9(1), 144-151. [https://doi.org/10.5539/ijel.v9n1p144]
- Semetko, H. A., & Valkenburg, P.M. (2000). Framing European politics: A content analysis of press and television news. Journal of Communication, 50(2), 93-109. [https://doi.org/10.1111/j.1460-2466.2000.tb02843.x]
- Seo, M. (2017, August 14). How does Yonhap News robot journalist ‘Soccerbot’ write a story. Yonhap News. http://www.yonhapnews.co.kr/bulletin/2017/08/14/0200000000AKR20170814032900039.HTML
- Shim, H. (2005). Examining narrative structure of Tsunami coverage of the Weekly Magazine 2580 by MBC and 60 Minutes by CBS. Korean Journal of Journalism & Communication Studies, 49(6), 286-313.
- Simpson, P. (2005). Language, ideology, and point of view. London, UK: Routledge.
- Teo, P. (2000). Racism in the news: A critical discourse analysis of news reporting in two Australian newspapers. Discourse & Society, 11(1), 7-49. [https://doi.org/10.1177/0957926500011001002]
- van Dalen, A. (2012). The algorithms behind the headlines: How machine-written news redefines the core skills of human journalists. Journalism Practice, 6(5-6), 648-658. [https://doi.org/10.1080/17512786.2012.667268]
- van Dijk, T. A. (1983). Discourse analysis: Its development and application to the structure of news. Journal of Communication, 33(2), 20-43. [https://doi.org/10.1111/j.1460-2466.1983.tb02386.x]
- Vincent, R. C. (2000). A narrative analysis of US press coverage of Slobodan Milosevic and the Serbs in Kosovo. European Journal of Communication, 15(3), 321-344. [https://doi.org/10.1177/0267323100015003004]
- White, P. R. R. (2000). Media objectivity and the rhetoric of news story structure. In E. Ventola (Ed.), Discourse and community: Doing functional linguistics (pp. 379-397). Tübingen, Germany: Gunter Narr Verlag.
- Yang, J., Jin, M., Lee, K., Oh, H., & Cho, J. (2017). User preference customized sports articles Writing program based on robot journalism, 1054-1056. Paper presented at the conference of Korea Information Science Society.
- Yoo, S., & Lee, G. (2017). A study on the tone difference among 4 general programming cable channels news programs – Content analysis of news articles about free welfare issue. Korean Journal of Journalism & Communication Studies, 61(1), 7-35.