Improved SVM with Hyperparameter Tuning for Fake News Detection

Abstract

In today’s digital age, accessing information has become effortless. An abundance of resources is available online, from trustworthy news outlets providing factual information to unverified opinions shared by anonymous individuals. With the advent of modern technology, social media platforms have revolutionized interaction and staying informed, providing instant access to news and information related to a wide range of topics. They also allow us to share valuable links and content that we find interesting or informative and express our thoughts and beliefs on various issues. However, knowing if the information you see is true or fake can be challenging. This study introduces an improved SVM with hyperparameter tuning for detecting fake news on the Twitter dataset. The proposed has two phases: Check-worthiness identification and fact-checking, which include three tasks: Feature selection, fake news detection and determining whether claims within tweets are factual. The main idea for tackling complex optimization problems is to transform them into more straightforward linear or quadratic programming problems. This transformation is made possible by approximating the Gaussian kernel using Epanechnikov kernels. The process involves selecting an optimal probability distribution from a set of choices and using the minimax strategy to construct the most effective separating functions. The approach is a highly efficient and effective way of addressing optimization problems that are too complex to solve through direct methods. According to the results, the proposed method has been able to identify fake news with accuracy, precision, recall and F-measure of 99.67, 99.61, 100 and 99.81%, respectively. This framework is a game-changer in the fight against misinformation, as it allows the classification of recurring fake news and the utilization of social network users’ connections to prevent the spread of false information.