Paper Type
ERF
Description
Bias in news media, expressed as ‘subjectivity’, exists (Lowry 2008) and can have serious consequences due media’s rapid digital dissemination. Techniques to identify media subjectivity range from human annotators to machine learning models (Recasens et al. 2013; Yano et al. 2010). However, most detection methods, from linguistics to deep-learning are ineffective, given the task complexity (Rodrigo-Ginés et al. 2024). Rewriting text using Large Language Models (LLMs) can mitigate subjectivity to objective news. This study examined the use of LLMs to rewrite news articles for this goal. The study used a dataset of 768 articles and TextBlob to measure subjectivity. Then, prompt engineering was used to rewrite articles with the aid of an LLM (Google’s Gemini Model). Text rewritten by the LLM showed significantly lower subjectivity compared to the original article. This research is the first step towards understanding how LLMs can be leveraged to minimize media subjectivity in news.
Paper Number
1675
Recommended Citation
Iyer, Neel and Ray, Deepa Kajal, "The Power of LLMs in Media Bias Mitigation" (2024). AMCIS 2024 Proceedings. 21.
https://aisel.aisnet.org/amcis2024/social_comp/social_comput/21
The Power of LLMs in Media Bias Mitigation
Bias in news media, expressed as ‘subjectivity’, exists (Lowry 2008) and can have serious consequences due media’s rapid digital dissemination. Techniques to identify media subjectivity range from human annotators to machine learning models (Recasens et al. 2013; Yano et al. 2010). However, most detection methods, from linguistics to deep-learning are ineffective, given the task complexity (Rodrigo-Ginés et al. 2024). Rewriting text using Large Language Models (LLMs) can mitigate subjectivity to objective news. This study examined the use of LLMs to rewrite news articles for this goal. The study used a dataset of 768 articles and TextBlob to measure subjectivity. Then, prompt engineering was used to rewrite articles with the aid of an LLM (Google’s Gemini Model). Text rewritten by the LLM showed significantly lower subjectivity compared to the original article. This research is the first step towards understanding how LLMs can be leveraged to minimize media subjectivity in news.
Comments
SOCCOMP