Scholarlink

<p>NLP has evolved from task-specific systems to broadly capable models across domains. Financial natural language processing (FNLP) is now an active area. This article presents a comprehensive, state-of-the-art review of (FNLP) approaches for automated extraction from corporate financial reports and sustainability disclosures. The advances in the field are surveyed from shallow and ensemble methods to deep learning and transformer-based large language models (LLMs). It assesses progress in understanding, extracting, and structuring information from complex, multi-format financial documents across languages. It covers methods for document-centric text and table extraction, narrative comprehension (exploring the role of sentiment and topic modeling), anomaly and risk detection, and question answering, alongside the tools and platforms that operationalize these capabilities. This review summarizes background knowledge on document structure, financial governance, and environmental narrative conventions and then presents an analysis of existing studies in chronological order that explore feature-engineered classifiers, ensembles, and multimodal, layout-aware, and knowledge-augmented neural models. Particular attention is given to multilingual and Arabic–English reporting contexts common in Gulf markets. Moreover, the review compares widely used extraction and analysis tools to guide practical selection for future studies. Additionally, FNLP applications are presented beyond core reporting, including compliance, auditing, sustainability analytics, and fraud screening, highlighting their growing significance for decision support. Finally, the review articulates open challenges and outlines future research directions to advance more accurate, transparent, and scalable FNLP.</p>

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

A Comprehensive Study on Automated Extraction from Financial Reports Using NLP Approaches

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Fawaz Khaled Alarfaj

摘要

NLP has evolved from task-specific systems to broadly capable models across domains. Financial natural language processing (FNLP) is now an active area. This article presents a comprehensive, state-of-the-art review of (FNLP) approaches for automated extraction from corporate financial reports and sustainability disclosures. The advances in the field are surveyed from shallow and ensemble methods to deep learning and transformer-based large language models (LLMs). It assesses progress in understanding, extracting, and structuring information from complex, multi-format financial documents across languages. It covers methods for document-centric text and table extraction, narrative comprehension (exploring the role of sentiment and topic modeling), anomaly and risk detection, and question answering, alongside the tools and platforms that operationalize these capabilities. This review summarizes background knowledge on document structure, financial governance, and environmental narrative conventions and then presents an analysis of existing studies in chronological order that explore feature-engineered classifiers, ensembles, and multimodal, layout-aware, and knowledge-augmented neural models. Particular attention is given to multilingual and Arabic–English reporting contexts common in Gulf markets. Moreover, the review compares widely used extraction and analysis tools to guide practical selection for future studies. Additionally, FNLP applications are presented beyond core reporting, including compliance, auditing, sustainability analytics, and fraud screening, highlighting their growing significance for decision support. Finally, the review articulates open challenges and outlines future research directions to advance more accurate, transparent, and scalable FNLP.