<![CDATA[This research addresses the challenges of summarizing long and complex legal documents, which often exceed the input length limitations of transformer-based models and contain intricate legal reasoning structures. The purpose of this study is to develop an efficient and scalable summarization framework that preserves semantic fidelity and structural coherence in judicial summaries. To achieve this objective, a hybrid summarization pipeline is proposed by integrating a Bidirectional Encoder Representations from Transformers (BERT)-based extractive model with a hierarchical abstractive model based on Distilled Bidirectional and Auto-Regressive Transformers (DistilBART), combined with a Divide-and-Conquer strategy. The proposed method partitions long legal documents into smaller segments, processes each segment independently, and reconstructs them into a coherent final summary. Experiments were conducted on the Indian Legal Case Summarization dataset and evaluated using Recall-Oriented Understudy for Gisting Evaluation (ROUGE), BERTScore, and Cosine Similarity to assess both lexical overlap and semantic similarity. The results show that the hierarchical DistilBART model outperforms the extractive baseline, achieving a ROUGE-1 score of 0.3802 and a Cosine Similarity of 0.6917. These findings demonstrate that the proposed framework provides an effective solution for long-document summarization in the legal domain.]]>
Copyrights © 2026
