South Korean newspaper coverage of Yemeni refugees: analysis of topics and sentiments using machine learning techniques

ABSTRACT

This paper aims to empirically investigate how South Korean newspapers define and report refugee issues. More specifically, we identify the prevalent topics and sentiments in the newspaper coverage of Yemeni refugees by using two machine learning techniques—structural topic model (STM) and Bidirectional Encoder Representations from Transformers (BERT). The analyses show that the most prevalent topic covered in the newspapers is ‘Humanitarian residence permit’—whether the government should provide it for humanitarian reasons—, followed by the topic ‘nationalism,’ which refers to criticism and concerns about losing ‘national identity’ by accepting more foreign residents. Hence, our results show that the local newspapers are more likely to report the need for humanitarian stay permits and convey factual information such as refugee crime, while the national newspapers tend to focus on contentious issues such as ‘nationalism.’ On the other hand, we find weak evidence for the difference in covered topics in Yemeni refugee news between conservative and liberal newspapers. The findings contribute to understanding how media frames refugee problems and also have policy implications.

KEYWORDS:

• Media
• Yemen
• Refugee
• Structural topic model (STM)
• Bidirectional encoder representations from transformers (BERT)
• Local newspapers

Acknowledgements

Joonseok Yang thanks the Korean Ministry of Education and the National Research Foundation of Korea (NRF-2020S1A3A2A02092791) for support for this research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Ministry of Justice in Republic of Korea Statistics, Accessed on December 26th, 2022 at https://kosis.kr/statHtml/statHtml.do?orgId=111&tblId=DT_1B040A6

2 The numbers of the news articles in each newspaper are summarized in Table A1.

3 As a robustness check, we also check the results by choosing the number of topics that give the second highest semantic values. We find that the results remain substantively same.

4 We utilized findThoughts function in the stm package to extract the most prevalent passages from documents.

5 To perform sentiment analysis of the news articles about Yemeni refugees, we utilized Naver CLOVA Sentiment platform that use the pre-training BERT model.

Additional information

Notes on contributors

Jaeyoung Hur

Jaeyoung Hur is an Associate Professor in Global Leaders College at Yonsei University. Professor Hur studies Public Opinion, Korean Politics, and Political Communication.

Joonseok Yang

Joonseok Yang is an Assistant Professor in the Department of Political Science & Diplomacy at Sungkyunkwan University, Seoul, South Korea. His research interests lie at the intersection of international political economy, foreign policy, public opinion, and quantitative and experimental methods.