FAKE NEWS DETECTION USING FINE-TUNEDPRETRAINED LANGUAGE MODELS

Abstract

Due to lightning growth of fake news the erosion of democratic and public trust, fake news detection and intervention are becoming increasingly important. In order to ensure that the public gets the correct information, various Machine Learning Deep Learming approaches have been developed regardless of whether the news is traditional or non traditional. The aim of this paper is to identify false news using cutting-edge detection methods i.e. Analysis of fake news taxonomy has been conducted, therefore the purposeis to examine cutting-edge methods for spotting false news and discuss the drawbacksof those methods. Additionally, it discussed credibilityBased, social contextBased, time Based substance-based detection of false news. In Addition we reviewed the recent work of other academics on the subject and provided a roadmap for future study to aid scholarsin combating the issue related to fake news And In recent years, proliferation of social media websites has resulted in a surge of fake news being created and disseminated for various commercial and political reasons. The use of misleading language in these online articles makes it easy for users to be misled, leading to significant offline consequences. Therefore, crucial steps should be taken to enhance the reliability of information in online social networks.we propose a technique for detecting fake-news that leverages fine-tuned pretrained language models(including BERT, AlBERT, DistilBERT, XLNet RoBERTa). Our approach involves preprocessing the text data, then fine tuning the pretrained models on dataset of labeled fake-real news, and then evaluating their performance on a held-outtest set. Our experiments demonstrate that finetuning pretrained models can effectively detect fake-news with high-accuracy, and can be a promising approach for building auto-mated systems to combat the spread of misinformation.