In this project, we have used a Siamese network to manipulate the base embedding on MPNet, such that fake news data and truth (news) are clustered separately and better distinguishable by classifier models. With such imporvement, we were able to maintain accuracy while we have a more dense and compressed embedding of the news.