Media Legitimacy Detection

The Project

Media sources, primarily of the political variation, have a hastening grip on narratives that can easily be constructed using biased views and false information. Unfortunately, many people in modern society are unable to differentiate these false narratives from real events. Utilizing natural language processing, sentiment analysis, and various other computer science techniques, models can be generated to help users immediately detect bias and falsehoods in political media. The models created in this experiment were able to detect up to 70% accuracy on political bias and 73% accuracy on falsehoods by utilizing datasets from a variety of collections of both political media and other mediums of information. Overall, the models were successful as the standard for most natural language processing models achieved only about 75% accuracy.

View this project on YouTube

Team Comments

The primary goal of the project is to explore whether data science can be adequately applied to a social science field of identifying political bias and falsehoods, specifically using natural language processing to correlate patterns in linguistics with false media.

Initial approaches utilized sentiment analysis, however, the primary problem in this usage is that political media can often have varying sentiment without being false. Ultimately, the project ended up using frequency pipelines to analyze the data, which did not rely on fake news being specifically

Future work would include the usage of a long term learning multilayer perceptron that would be trained by concatenation of multiple successful datasets. This standardizes the datasets further by using the exact same model and a corroboration of all datasets could compensate for weakness in each one