• Technical Name
  • Using Artificial Intelligence to Build a Safe and Secure Social Media System
  • Operator
  • Academia Sinica
  • Booth
  • Online display only
  • Contact
  • 邵雅玲
  • Email
  • angelshao@mail.ntust.edu.tw
Technical Description Disinformation proliferation is a significant issue in recent years. From jeopardizing mental health to provoke panic during an epidemic, it is clear that disinformation is a pressing issue. Our government and NGO have thus erected various online platforms for information clarification and verification, and our goal is to support them through our research.
In our research, we noticed that the spread of disinformation heavily relies on people reposting those articles through the internet. We believe if we can act as the mediator between articles and the people reading them, we will be able to reduce disinformation propagating significantly. Therefore, we construct a system using AI, NLP, searching engine technology, and unstructured databases to examine any given articles. This system provides a channel to supervise the background information of the articles, which will help people improve their media literacy and stopping them from reposting disinformation.
Scientific Breakthrough For our data, we collect various articles and posts from Taiwan's mainstream media, including press and social media platforms. For our model, since most of the relevant research is designed for English only, we had to collect the Chinese corpus data and complete the corresponding semantic analysis ourselves. Our Clickbait Model and Title and Content Similarity Model show the excellent results in media literacy of the news in Taiwan.
Industrial Applicability Most of the existing domestic disinformation prevention mechanisms are done through manual review. Though effective, manual inspection is way to slow to catch up with the rapidly growing disinformation. By supplementing the background information of the received messages, we help people understand the reliability of the information they got and reduce their willingness to repost disinformation. This not only avoids labeling and political problems, but it also prevents potential violations of freedom of speech.