Building a natural language processing (NLP) model for effective legal & financial document summarization

Abstract

Digital judicial judgement papers allow for data extraction and application. Due to their peculiar structure and complexity, automatic summarizing of these legal writings is vital and difficult. Previous techniques have used large labeled datasets, hand-engineered features, domain expertise, and a small sub-domain for greater efficacy. We offer simple generalized neural network summarizing methods for Indian court judgment papers in this study. Two neural network designs using sentence and word embeddings for semantics are examined. The suggested methodologies may be used to different domains since they do not need hand-crafted features or domain-specific expertise. We award classes/scores to phrases in the training set based on their match with human-produced reference summaries to address the lack of labeled data. Our suggested methods outperform other baselines in experimental assessments.