Numerous enhancements have been made to the mobile internet, which has led to an increase in people?s attention to posting more multi-modal posts among social media platforms. Detecting fake news continues to be a serious concern in today's social culture. Machine learning and deep learning algorithms have shown significant results in predicting and identifying fake news. In recent studies, existing deep learning 1-D CNN models have proved very effective for text classification problems. Due to a large number of floating-point operations, a large number of layers with many hidden units necessitates an enormous amount of memory space and a high response time. Today, the number of edge devices is growing exponentially, making it difficult to deploy CNN models that have been pre-trained directly on these devices. This paper addresses the compression issue by proposing a particle swarm-based optimization technique to retain the best-hidden units without compromising the classification model's accuracy. In this paper, we have chosen Top-1 accuracy and compression ratio as the dual objective fitness functions, so that accuracy should not degrade excessively after compression. The results of these experiments are examined using the proposed methodologies. The analysis of the results reveals positive performance with the use of the proposed deep neural network.