<p>The rapid growth of online news websites complicates the identification of trending topics characterized by volume, diversity, and velocity. Traditional classification methods struggle with this complexity, necessitating the use of advanced techniques to address it. In this study, a Fusion clustering framework that combines DBSCAN and Gaussian Mixture Models (GMM) to improve scalability and accuracy in hot news topic recognition is introduced. The proposed approach employs TF-IDF (Term Frequency-Inverse Document Frequency) in extracting and pre-processing features, such as tokenisation, punctuation elimination, and stop word elimination. Classification is performed with the help of the model that is a combination of Gated Recurrent Unit (GRU) networks and Bi-LSTM networks and is capable of handling the temporal dependencies of news data. The model employs parallel operating with a cloud computing structure to enhance greater efficiency and optimisation of large datasets. Based on the results of the tests, we discover that the model classification performance has 97.39% accuracy, 0.97 precision, and 96.77 sensitivity, and thus reports news categories with a high degree of accuracy. Besides, the model has the ability to counter the real-world challenges of news topic discovery, such as noise removal, multi-source data integration and real-time processing. The proposed solution is an effective way of addressing the issue of large-scale news websites that would need to enhance topic discovery and accuracy of classification.</p>