Scholarlink

<p>The study examines the opinions expressed in 143,000 reports from 15 different U.S. news sources between 2000 and 2017 using cutting-edge Natural Language Processing (NLP) approaches, such as Recurrent Neural Networks with Long Short-Term Memory units and Transformer-based models like BERT and GPT. Unlike previous studies that focus on short-term sentiment analysis or limited sources, our comprehensive dataset spans nearly two decades and encompasses a diverse range of publications, capturing evolving emotions and media biases over time. Web scraping and intensive preprocessing were used to carefully curate the dataset, which captured the changing emotions over time across a variety of publications. The sentiment analysis shows that media coverage is generally biased in a positive way, with notable variations that correspond to important world events. Additionally, the study reveals significant emotional variation among news organizations, which reflects their distinct editorial stances and target audiences. The emotional tone of articles is also clearly influenced by the individual authors, highlighting the part that individual writing styles have in influencing public opinion. Furthermore, linguistic diversity and sentiment expression are found to be correlated, indicating that more complex emotional tones may be linked to a diverse vocabulary. Consistent changes in emotion over time are shown by the temporal analysis, which corresponds with sociopolitical and economic advancements. These results show how important media are in reflecting public sentiments and how well NLP methods work to glean insightful information from massive amounts of textual data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Analyzing two decades of media sentiments: NLP and deep learning insights into news bias and trends

  • 期刊论文

  • 发表日期：

    2025年2月17日

查看全文
  • Shardha Purohit,
  • Vijayta Taneja,
  • Palak Saxena,
  • Rusha Mudgal,
  • Pooja Rana,
  • Ritu Garg,
  • Owais Ahmad Shah

摘要

The study examines the opinions expressed in 143,000 reports from 15 different U.S. news sources between 2000 and 2017 using cutting-edge Natural Language Processing (NLP) approaches, such as Recurrent Neural Networks with Long Short-Term Memory units and Transformer-based models like BERT and GPT. Unlike previous studies that focus on short-term sentiment analysis or limited sources, our comprehensive dataset spans nearly two decades and encompasses a diverse range of publications, capturing evolving emotions and media biases over time. Web scraping and intensive preprocessing were used to carefully curate the dataset, which captured the changing emotions over time across a variety of publications. The sentiment analysis shows that media coverage is generally biased in a positive way, with notable variations that correspond to important world events. Additionally, the study reveals significant emotional variation among news organizations, which reflects their distinct editorial stances and target audiences. The emotional tone of articles is also clearly influenced by the individual authors, highlighting the part that individual writing styles have in influencing public opinion. Furthermore, linguistic diversity and sentiment expression are found to be correlated, indicating that more complex emotional tones may be linked to a diverse vocabulary. Consistent changes in emotion over time are shown by the temporal analysis, which corresponds with sociopolitical and economic advancements. These results show how important media are in reflecting public sentiments and how well NLP methods work to glean insightful information from massive amounts of textual data.