LSSS-System: Fake Content Detection Using Linguistic Features

In our study, we employed web scraping techniques to collect data from PolitiFact.com, a reputable fact-checking platform maintained by the nonprofit Poynter Institute, known for distinguishing political fiction from fact. We analyzed over 6,000 political, celebrity, financial, and economic statements using linguistic features such as syntactic, sentiment, stylistic, and lexical aspects to differentiate between authentic and fraudulent information. Machine learning models were used to extract and evaluate these features to identify which ones were most predictive for detecting fake and real content.Our analysis revealed that syntactic features, like punctuation count and verb count, had the highest positive importance, meaning they significantly contributed to the model’s accuracy. However, features like sentiment polarity, vocabulary richness, and noun count exhibited negative importance, indicating they may introduce noise or reduce the model’s performance. By identifying which features positively or negatively impacted the model, we refined our detection framework to enhance the reliability of fake content classification.