Developing and examining hybrid classifiers to study social media fake news

Abstract The proliferation of technological innovations in realm of social media has facilitated ability of individuals to articulate and disseminate their viewpoints. The consensus reached may exhibit prejudice, inaccuracy, or potential for misrepresentation. Frequently, society is drawn to erroneous information due to its allure, despite potential dangers and harm it may pose. To enhance credibility, these communications consist of textual content, visual elements, and/or hyperlinks. Rich communications possess the capacity to alter the opinions of individuals or groups, perhaps resulting in deterioration of their individual or collective image and reputation. This study presents a novel hybrid context-aware approach that incorporates implicit information and contextual factors in classification of text as deceptive. The viability of our proposed model was assessed using three benchmark datasets. The findings yielded persuasive and encouraging outcomes across all datasets. Additionally, performance metrics (accuracy, precision, recall, specificity, f1-score, AUC, and l-measure) of suggested models were compared to that of the baseline models (RNN, LSTM, and GRU), while requiring fewer compute resources, less time, and lower cost. Business strategies have been developed utilizing implicit data and contextual factors extracted from text to make predictions on the authenticity or falseness of information. Many of proposed methods only capture a portion of the relevant background, resulting in predictions that are not entirely accurate.