Multi-document summarization using LLAMA2 model with transfer learning

This paper presents an advanced framework for multi-document summarization (MDS) using LLaMA2, a state-of-the-art large language model developed by Meta AI. The proposed approach leverages transfer learning, sentence embeddings, and clustering techniques to generate coherent, concise, and high-quality summaries from large volumes of heterogeneous documents. By chunking input text, extracting semantic vectors, and applying K-Means clustering, representative content from each group is identified and summarized using LLaMA2, which is then further refined into a final abstractive summary. Evaluation results demonstrate that the LLaMA2-based model outperforms other models such as FLAN-T5 and FLAN-UL2 across key performance metrics including ROUGE, BLEU, METEOR, coherence, and readability. The system's versatility allows it to be deployed across diverse domains such as medical reports, research literature, and multilingual news datasets. This work underscores the effectiveness of LLaMA2 in addressing the core challenges of MDS—such as duplication, long context, and semantic diversity while offering a scalable and efficient solution for real-world summarization tasks.