Frog-Leaping Search-Optimized BiLSTM-Attention Network with GNN for News Hotspot Prediction and Dissemination Path Optimization

Abstract

The exponential growth of online news and social media platforms necessitates intelligent systems capable of predicting emerging news hotspots and optimizing dissemination pathways. This research proposes a novel deep learning framework, Frog-Leaping Search-Mutated BiLSTM with Attention Net (FLS-BiLSTM AttNet), designed to identify potential news hotspots and enhance information propagation accurately. The dataset was collected from multiple reliable sources, including the Kaggle repository, Google News API, Twitter feeds, and various online news portals. Preprocessing involves removing stop words, punctuation, and irrelevant symbols, followed by tokenization to convert text into meaningful word sequences. TF-IDF is applied to extract significant keywords, quantifying their relevance within individual documents and across the dataset. The Bidirectional Long Short-Term Memory (BiLSTM) network captures temporal dependencies in news sequences, while the Attention Net (AttNet) highlights critical features for improved prediction. Frog-Leaping Search (FLS) optimizes weight initialization, enhancing convergence speed and hotspot detection performance. To optimize dissemination pathways, Graph Neural Networks (GNNs) are employed to identify efficient propagation routes, reducing latency and maximizing coverage. Implemented in Python, the model achieves an accuracy of 98.92%, a precision of 98.45%, a recall of 98.41%, an F1 score of 98.57%, a coverage of of of 94.5%, and an NDCG of of of 82.6%, demonstrating substantial improvement over traditional methods. The proposed framework successfully integrates temporal modeling, attention mechanisms, optimization strategies, and graph-based propagation, making it highly applicable to real-world digital journalism, early warning systems, and media strategy planning, while maintaining robustness and adaptability across diverse news datasets.