Fake News Detection with Credibility Signals

Abstract

In the era of huge information, it is necessary to justify the online contents whether it is true or false. Nowadays it is a great challenge to detect false information as the online contents sometimes show numerous misinformation. The proposed model in troduces a method to detect misinformation significantly to generate different credi bility signals that ensures the truthfulness and authenticity of the content. It involves advanced computational requirements and machine learning algorithm techniques that specify online contents focusing on the extraction of credibility signals to en hance the credibility with reliable sentiment of the content. This method proposes a novel approach that leverages 8 different distinct credibility signals. This method in novatively utilizes the credibility signals like Emotional Violence, Incorrect Spelling, Evidence, Source Credibility, Polarized Language, Bias, Writing Quality, and Contra diction of Established Facts. Our proposed methodology works upon use as a fact checker with high quality and efficiency on the Politifact dataset which contains more than 21,000 unique statements that are also verified. The extraction of the credibility signals makes it more complex to determine the statement with its robustness. So our proposed model introduces a comprehensive model pipeline to improve the adaptabil ity of fake news from several online contents. We also made an experimental study on comparing the performance of the model with several pre-trained models like BERT, RoBERTa, XLNet, AlBERT, and ChatGPT. The proposed model shows a significant improvement in model accuracy and F1 score that makes the model superior repre sentation in detecting fake news. Regardless of these progressions, the study high lights challenges to extract the credibility signals. Future research aims to integrate more credibility signals to enhance the models performance and explainability by also working on multilingual perspects. The development of the model introduces an ef fective prompting to detect the trustworthiness of the online contents to give more accurate predictions.