- 디지털인문사회과학부 스페셜세미나-(9/3 수. 16:00), Ideological Signaling and Cross-Ideological Interactions on Social Media
- 관리자 |
- 2025-08-18 17:20:18|
- 23
- 2025-08-18 17:20:18|
디지털인문사회과학부에서 특별 세미나를 개최합니다. 관심 있는 분들의 많은 참여 바랍니다.
Title: Ideological Signaling and Cross-Ideological Interactions on Social Media
Speaker: Byungkyu Lee (New York University)
Date / Time: Sep 3rd, 2025 (Wed) 4pm-5pm
Venue: N4-1433
Inquiry: lanukim@kaist.ac.kr
Abstract:
Social media platforms are often conceptualized as either echo chambers that amplify like-minded opinions or as public spheres fostering diverse interactions. Despite extensive research, evidence on the extent of cross-ideological exposure remains mixed. Previous studies primarily rely on interactional data to gauge cross-ideological engagement yet often overlook the nuanced ideological signaling users express in cross-ideological interaction. To address this gap, we develop a novel methodological framework using large language models to quantify ideological signaling. Leveraging data from hundreds of millions of comments posted on Facebook’s public forums during the 2015–2016 U.S. election period, we first estimate users’ ideology based on their liking patterns across about 500 major partisan media pages and then fine-tune BERT to predict ideological alignment from users' comments alone. Our findings reveal a significant degree of ideological signaling, even within nonpolitical discussions, allowing accurate differentiation between liberals and conservatives. Additionally, we examine the contexts and mechanisms that amplify ideological signaling, showing that users adjust their ideological cues in response to cross-ideological interactions. This study highlights the central role of ideological signaling in understanding online polarization and demonstrates a scalable, language model-based approach for measuring it across digital platforms.
Title: Ideological Signaling and Cross-Ideological Interactions on Social Media
Speaker: Byungkyu Lee (New York University)
Date / Time: Sep 3rd, 2025 (Wed) 4pm-5pm
Venue: N4-1433
Inquiry: lanukim@kaist.ac.kr
Abstract:
Social media platforms are often conceptualized as either echo chambers that amplify like-minded opinions or as public spheres fostering diverse interactions. Despite extensive research, evidence on the extent of cross-ideological exposure remains mixed. Previous studies primarily rely on interactional data to gauge cross-ideological engagement yet often overlook the nuanced ideological signaling users express in cross-ideological interaction. To address this gap, we develop a novel methodological framework using large language models to quantify ideological signaling. Leveraging data from hundreds of millions of comments posted on Facebook’s public forums during the 2015–2016 U.S. election period, we first estimate users’ ideology based on their liking patterns across about 500 major partisan media pages and then fine-tune BERT to predict ideological alignment from users' comments alone. Our findings reveal a significant degree of ideological signaling, even within nonpolitical discussions, allowing accurate differentiation between liberals and conservatives. Additionally, we examine the contexts and mechanisms that amplify ideological signaling, showing that users adjust their ideological cues in response to cross-ideological interactions. This study highlights the central role of ideological signaling in understanding online polarization and demonstrates a scalable, language model-based approach for measuring it across digital platforms.
| 첨부파일 |
|---|
- 다음
- 장정우 교수 논문, Energy & Environment 저널에 출판
- 2025-08-07