DEVELOPING AN NLP MODEL FOR EFFICIENT SUMMARISATION OF LEGAL AND FINANCIAL DOCUMENTS

Abstract

The rapid growth of digital legal and financial documentation has created a significant demand for intelligent summarization systems capable of extracting essential information while preserving contextual accuracy. Automated text summarization has become increasingly important for improving document accessibility, accelerating decisionmaking, and reducing the time required for manual analysis across professional domains. Conventional summarization approaches often struggle to capture domain-specific terminology, complex sentence structures, and contextual dependencies, leading to incomplete or less informative summaries. A domain-specific corpus containing legal contracts, court judgments, financial reports, and regulatory documents collected from publicly available repositories between 2018 and 2025 is utilized, incorporating diverse textual features such as document length, semantic relationships, named entities, and contextual embeddings. The collected data undergoes preprocessing through text normalization, tokenization, stop-word removal, lemmatization, sentence segmentation, and transformer-based encoding to improve representation quality. Multiple transformer architectures, including BERT, RoBERTa, T5, BART, and PEGASUS, are implemented and compared with a hybrid TransformerAttention framework for abstractive summarization. Performance is assessed using ROUGE-1, ROUGE-2, ROUGEL, BLEU, and BERTScore metrics, where the hybrid Transformer-Attention model achieves the highest performance with a ROUGE-L score of 0.921, BLEU score of 0.874, and BERTScore of 0.947, producing concise and contextually consistent summaries. The proposed framework significantly enhances summarization quality, semantic preservation, and domain-specific information retention for legal and financial documents.