Scholarlink

Long-text summarization, particularly for books, remains a key challenge in natural language processing. With the growth of academic and educational materials, efficient summarization is increasingly vital. This work introduces a hybrid method combining extractive and abstractive techniques, alongside purely abstractive models for a comprehensive evaluation. It presents a hybrid extract – then – abstract approach evaluated on BookSum using LLaMA-3-8B-Instruct and Qwen2.5-7B-Instruct under four settings: (i) base abstractive, (ii) fine-tuned abstractive (LoRA/QLoRA), (iii) Hybrid – Base, and (iv) Hybrid – Fine-tuned. The Hybrid – Fine-tuned LLaMA achieves the best ROUGE and BERTScore results. Compared to recent models such as NexusSum and TopDownFormer, our system attains competitive ROUGE and higher BERTScore, demonstrating that extractive guidance enhances coherence while preserving content. Overall, it offers an efficient and practical strategy for long-text summarization.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Long-Text Summarization Through Hybrid Extractive-Abstractive Methods with LLaMA and Qwen

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mostafa Magdy,
  • Salwa O. Slim,
  • Ensaf Hussein Mohamed

摘要

Long-text summarization, particularly for books, remains a key challenge in natural language processing. With the growth of academic and educational materials, efficient summarization is increasingly vital. This work introduces a hybrid method combining extractive and abstractive techniques, alongside purely abstractive models for a comprehensive evaluation. It presents a hybrid extract – then – abstract approach evaluated on BookSum using LLaMA-3-8B-Instruct and Qwen2.5-7B-Instruct under four settings: (i) base abstractive, (ii) fine-tuned abstractive (LoRA/QLoRA), (iii) Hybrid – Base, and (iv) Hybrid – Fine-tuned. The Hybrid – Fine-tuned LLaMA achieves the best ROUGE and BERTScore results. Compared to recent models such as NexusSum and TopDownFormer, our system attains competitive ROUGE and higher BERTScore, demonstrating that extractive guidance enhances coherence while preserving content. Overall, it offers an efficient and practical strategy for long-text summarization.