THESIS
2021
1 online resource (xiii, 152 pages) : illustrations (some color)
Abstract
Fake news is capturing increasing attention worldwide and severely impacting our society.
In this thesis, I examine three aspects of this issue with the help of computerized textual analysis
techniques. In the first essay, I propose a domain adaptive transfer learning approach to detect fake
content. I extract domain-invariant linguistic features associated with fake general news and
transfer them to three specific domains: political news, financial news and online reviews, to
address the missing label problem in fake news detection. I further derive a measure to explain the
utility of transfer learning. In the second essay, I investigate the impact of fake news on finance
and social media. For the impact on finance, I first show that humans cannot tell the authenticity
of news, while t...[
Read more ]
Fake news is capturing increasing attention worldwide and severely impacting our society.
In this thesis, I examine three aspects of this issue with the help of computerized textual analysis
techniques. In the first essay, I propose a domain adaptive transfer learning approach to detect fake
content. I extract domain-invariant linguistic features associated with fake general news and
transfer them to three specific domains: political news, financial news and online reviews, to
address the missing label problem in fake news detection. I further derive a measure to explain the
utility of transfer learning. In the second essay, I investigate the impact of fake news on finance
and social media. For the impact on finance, I first show that humans cannot tell the authenticity
of news, while their perceived fakeness of financial news influences their trading intention. I then
conduct a survey study to examine what factors constitute fake news perception. By leveraging
data collected from the survey study, I apply domain adaptive transfer learning to infer any
financial news’s perceived fakeness and show that this deep learning-based perceived fakeness
predicts less trading volume and stock volatility. For the impact on social media, I apply the same
model to study how individuals interact with deceptive tweets on social media during social crises
and how they respond to these tweets. In the third essay, I investigate the effectiveness of two
platform interventions (i.e., a fake news flag and a forwarding restriction policy) in combating fake
news. By leveraging latent semantic analysis, word lexicon, and topic modeling, I provide
empirical evidence to the impacts of platform interventions on fake news dissemination and
survival, as well as the underlying mechanisms.
Post a Comment