Project

NEWS DIVING: News Sentiment Diversity, Index construction and Generative models

Code
3G010920
Duration
01 January 2020 → 31 December 2023
Funding
Regional and community funding: Special Research Fund, Research Foundation - Flanders (FWO)
Promotor
Research disciplines
  • Social sciences
    • Knowledge management
Keywords
Highfrequency Textual analysis Time series
 
Project description

The digitalization of the media sector has disrupted the access to news News articles are now available at a large scale and low cost across many news sources In spite of the increase in news volume, economic agents often only consume news provided by a non-diversified subset of news sources The lack of diversification reveals itself in both the choice of topics covered and the sentiment used in the news articles This may lead to a biased view on the state of the economy and inefficient decision making  This project contributes insights and tools helping economic agents to increase the diversity and hence the efficiency of their news-based decisions in a digital world We have three research questions First, we investigate the problem of synchronisation and measurement of dependence of article-level news sentiment when the same events are discussed by different news sources at different times and in different languages Second, we study how to aggregate article-level sentiment into diversified news sentiment indices and use them in sentiment diffusion models that describe the dynamic relationship between news diversity and the economic and financial system Third, we explore the use of generative models and regime switching to evaluate the impact of state variables on the multivariate distribution of sentiment For each research question, we dive into the archives of millions of news articles and provide new insights about news sentiment diversity