Hybrid Entropy-based Weighted Graph Embedding Model for Multi-Document Text Summarization

Managing the ever-growing volume of textual information presents a pressing challenge in today's information-rich landscape. Multi-document summarization, a vital task in this context, faces formidable hurdles such as redundancy, information misalignment, and the critical need for context preservation. Recent advancements in natural language processing have introduced innovative techniques to tackle these challenges. Hence this research delves into these challenges and proposes an innovative solution the Hybrid entropy-based weighted graph embedding model based Elytron Vespid-B optimized BiLSTM model (HEWGEF+EVR BiLSTM). The objective is to significantly advance multi-objective multi-document text summarization techniques. These challenges are solved by employing standard methods for preprocessing multi-document data. The TF-IDF vectors are utilized to extract word significance, a fundamental step in content evaluation. The novel approach incorporates the Hybrid entropy-based weighted graph embedding model, which evaluates sentence-word correlations through entropy-based attribute elimination. This unique combination results in feature-rich representations of the text. The summarization process is enhanced with the Rank-based optimized BiLSTM classifier, equipped with a dedicated layer for sentence prioritization. This strategic approach optimizes sentence ranking, a pivotal aspect of effective summarization. Notably, the parameters of the BiLSTM classifier are optimally tuned using collaborative swarm optimization, merging wasp and beetle behaviors for optimal performance. The experimental results underscore the effectiveness of the proposed method across diverse sources. For the Daily Mail dataset (d1), achieved F1-scores (ROUGE-1), F1-scores (ROUGE-2), and F1-scores (ROUGE-L) of 0.70, 0.39, and 0.67, respectively. Similarly, for the DUC 2002 dataset (d2), the proposed model obtained F1-scores (ROUGE-1) of 0.73, F1-scores (ROUGE-2) of 0.35, and F1-scores (ROUGE-L) of 0.72. Likewise, for the DUC 2004 dataset (d3), the proposed model secured F1-scores (ROUGE-1) of 0.65, F1-scores (ROUGE-2) of 0.34, and F1-scores (ROUGE-L) of 0.67.