|The researchers' AI model would cement the study of additional linguistic features / Photo Credit: Panchenko Vladimir (via Shutterstock)|
Differentiating fake news from satire comes down to semantic and linguistic differences, but it can be difficult to spot, according to Kyle Wiggers of VentureBeat, a tech news website. Hence, researchers at George Washington University, Amazon AWS AI, and startup AdVerifai resorted to employing a machine learning approach to classify misleading speech. Or Levi and colleagues claimed the AI model they developed would serve as a foundation studying additional linguistic features.
The authors’ research followed MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) work earlier this year. CSAIL created an AI model that could determine if a source is accurate or politically prejudiced. Levi and colleagues’ study observed that efforts to curb the spread of misinformation have occasionally led to flagging legitimate satire on social media. To make things complicated, some fake news purveyors masquerade themselves as satire websites.
The researchers conjectured that “metrics of text coherence might be useful in capturing semantic relatedness between sentences of a story.” Coh-Metrix, a tool that produces linguistic and discourse representations implemented a set of indices related to text statistics, which were used by the researchers. They then employed a statistical technique called principal component analysis “to convert potentially correlated metrics into uncorrelated variables.” They used the variables in two logistic regression models together with the fake and satire labels.
Levi and his team analyzed the models’ performance by using 283 fake news and 203 satirical stories on a corpus, which were all verified by hand. The algorithm scored 0.78, where 1 indicates a perfect score. The results revealed that satirical news tended to sound more sophisticated “and less easy to read” than fake news articles.
The authors planned to study linguistic cues like incongruity, absurdity, and other humor-related features. They wrote that with the improved understanding of nuances between fake news and satire and classification accuracy, their contributions “carry great implications with regard to the delicate balance of fighting misinformation while protecting free speech. ”