Enhancing the Fake News Detection by Applying Effective Feature Selection Based on Semantic Sources

Abstract

Capturing reliable information from social networks is a challenge due to fake news risks. Existing works face shortages in exploiting short text processing, and in utilizing semantic-based resources to select optimal features. This paper proposed a CNIRI-FS (Contextual Negation Handling and Inherent Relation Identification for Enhanced Feature Selection) model to detect fake information; utilizing Wikipedia to add semantic features and an external enrichment from trusted web pages. A Genetic Algorithm (GA) was used to filter out unreliable features. The optimal feature set along with the negation handled features is validated using machine learning classifiers. The CNIRI-FS model results showed higher precision and accuracy than a model without optimal feature selection.