Today's internet users utilize social media sites extensively to share news and ideas. Social media is increasingly popular for communication because, unlike traditional media, it is accessible almost anywhere and anytime due to the extensive usage of smartphones. Fake news has proliferated on social media for societal or personal benefit. Most previous research works mostly use data that is taken directly from the news articles. This paper focuses on the news article's substance as well as the presence of echo chambers in the social network to detect false news. The pre-processing, news-user engagement matrix, and fake news classification are the primary work of this effort. Using a linked matrix–tensor factorization method, the proposed method fuses the news content with the tensor. A matrix formed after fusing the content is taken as input matrix for Deep Belief Network (DBN) for classifying the news articles as fake or real. The DBN is an ordered stack of numerous Restricted Boltzmann Machines (RBM) that functions as a probabilistic model. A DBN learning process comprises phases for both supervised and unsupervised learning. During the DBN pre-training phase, the Contrastive Divergence (CD) technique is employed to produce network weight parameters. This approach is layer-by-layer and greedy in an unsupervised manner. Next, a supervised Back Propagation (BP) method is used to adjust the weight levels. PolitiFact and BuzzFeed-based fake news datasets were used to validate the suggested approach's performance in terms of precision, recall, f-measure, and accuracy.