Project Team
Students
Ryan Himes
Computer Science
Penn State Harrisburg
Faculty Mentors
Truong Tran
Penn State Harrisburg
Computer Science/Mathematics
Project
Project Video
Project Abstract
Misinformation, or fake news has become a significant problem in the modern day. This can lead to changing peoples’ world views that revolve around information that is not true. Hence, the importance of automatically detecting it has become necessary with the amount of news freely flowing throughout the internet. Companies and researchers in the past have found ways to reliably detect whether or not news contains misinformation in it, or not, through binary classification methods. Sometimes this classification cannot tell the overall story of how truthful certain news is when parts of the news are truthful while the other parts are not. This problem can be fixed by applying a more precise label than true or false to a story, such as mostly true or half
true. To do this more accurately, a hierarchical ensemble method is proposed, which follows a divide and conquer approach that splits the data into smaller sub-sets to label news more accurately. After testing on three different news data sets, the hierarchical ensemble method outperforms all other known methods on two out of the three data sets. The only method that exceeds the hierarchical method on one data set is the most robust known method, stacking, and only by .22%. These results show that the hierarchical model can be a viable ensemble method for other data sets.
Evaluate this Project
Use this form link to provide feedback to the presenters, and add your project evaluation for award(s) consideration.