Mobile Text Misinformation Identification Using Machine Learning

Abstract

More than eighty percent of U.S. adults receive news from digital devices like smartphones, computers, or tablets. Unlike the traditional news dominated by organizations, this new kind of news could be created by anyone. It is quick and engaging. At the same time, misinformation may be easily generated or spread intentionally or unintentionally. Misinformation is a serious problem for the general public, and there is no method to solve the problem satisfactorily so far. Instead of covering general misinformation, this research tries to identify mobile health text misinformation by proposing a self-reconfigurable system. The system includes the preprocessing functions (involving lexical analysis, stopword removal, stemming, and synonym discovery), a dataflow graph from TensorFlow, and a reconfiguration method for self-improvement. Experiment results show the proposed method significantly improves the accuracy of the mobile health text misinformation detection compared to the one without using self-reconfiguration.