Within a second on social media, millions of posts were shared by people across multiple domains. Social media platforms such as \(\mathbb {X}\) , Facebook, Instagram, WhatsApp, Weibo, and others offer simple access and use. People share posts about their hobbies, entertainment, healthcare, home care, political information, and much more on these sites without fact-checking. After spreading this information, many people followed and propagated it. As a result, many problems are created, and society suffers. So, we need to detect these false or misinformation on social media. This research introduces a novel UcConvoNet model for detecting fake news. Firstly, the model extracts linguistic patterns from the text field of news articles. Next, it generates news article embedding representations, and these feature matrices are then concatenated into one matrix and fed into the proposed deep Convolutional Neural Network (CNN) for fake news classification. The algorithm learns features from news stories based on content and context, which is accomplished by concatenating composite features with CNN. In this work, we have used four real-time datasets of different sizes and one synthetic dataset, “CLUB-Fake,” which is a combination of all four datasets. “CLUB-Fake” resolves the issues of bias and large-scale dataset availability. The proposed model achieved a higher accuracy of 93.44% on McIntire, 99.27% on Kaggle, 99.47% on Reuter, 75.67% on BuzzFeed, and 97.72% on CLUB-Fake datasets than state-of-the-art works.
