Title: Independent Component and Vector Analyses for Explainable Detection of Misinformation During High Impact Events
Speaker: Zois Boukouvalas, American University, Dept. of Mathematics and Statistics
Date: Tuesday March 23, 2021
Abstract: With the evolution of social media, cyberspace has become the de facto medium for users to communicate during high impact events such as pandemics, natural disasters, terrorist attacks, and periods of political unrest. However, during such high impact events, misinformation in social media can rapidly spread, affecting decision making, creating social unrest, as well as creating financial instability. This has been observed in the present scenario with COVID-19. Identifying and curtailing the spread of misinformation during such a high impact event is a significant data challenge given the variety of data, as well as the speed by which misinformation can propagate. Recent machine learning advances have shown promise for the detection of misinformation, however, there are still key limitations that make this a significant challenge. These limitations are related to the use of multi-modal data as well as the explainability and reliable performance of a system geared at the detection of misinformation. In this talk, we discuss a data-driven solution that is based on Independent Component Analysis (ICA), such that detection of misinformation and knowledge discovery can be achieved jointly. In addition, we present a framework that is based on Independent Vector Analysis (IVA) to explicitly learn mutual relationships among different data modalities by letting multiple sources of information such as text, images, meta-data to interact adaptively while generating the joint representations for misinformation detection.