Paper: arXiv 2502.08875

Authors: Hsin-Min Lu, Yu-Tai Chien, Huan-Hsun Yen, Yen-Hsiu Chen

Abstract

Extracting specific items from 10-K reports is challenging due to variations in document formats and item presentation. To improve over traditional rule-based approaches, this study introduces and compares two advanced item segmentation methods: (1) GPT4ItemSeg, using a novel line-ID-based prompting mechanism to utilize a large language model, ChatGPT-4o, for item segmentation, and (2) BERT4ItemSeg, combining a pre-trained language model, BERT, with a Bi-LSTM model in a hierarchical structure to overcome context window constraints. Trained and evaluated on 3,737 annotated 10-K reports, BERT4ItemSeg achieves a macro-F1 of 0.9825, surpassing GPT4ItemSeg (0.9567), conditional random field (0.9818), and rule-based methods (0.9048) for core items (1, 1A, 3, and 7). These approaches enhance item segmentation performance, improving text analytics in accounting and finance. BERT4ItemSeg offers satisfactory item segmentation performance, while GPT4ItemSeg can easily adapt to regulatory changes. Together, they provide an extensible framework for 10-K item segmentation that supports reliable and reproducible results.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper provides a highly practical and empirically rigorous solution to a major financial text-processing bottleneck (10-K item segmentation) using state-of-the-art NLP techniques. While the mathematical complexity is low, the empirical methodology, benchmark creation (3,737 annotated filings), and novel line-ID-based prompting mechanism (LIB) offer substantial practical value and high reproducibility for computational accounting research.

Research Flowchart

  flowchart TD
    A["Research Goal: Improve 10-K Item Segmentation over Rule-Based Methods"] --> B["Data Input: 3,737 Annotated 10-K Reports"]
    
    B --> C1["BERT4ItemSeg: Hierarchical BERT + Bi-LSTM<br/>(Overcomes context window limits)"]
    B --> C2["GPT4ItemSeg: ChatGPT-4o<br/>(Line-ID-based prompting)"]
    
    C1 --> D["Evaluation & Comparison<br/>(vs. Rule-based & CRF baselines)"]
    C2 --> D
    
    D --> E1["Key Finding: BERT4ItemSeg achieves top Macro-F1 (0.9825)"]
    D --> E2["Key Finding: GPT4ItemSeg (F1: 0.9567) offers easy adaptability"]

Cited References (44)

  1. The Development of a RAG-Based Artificial Intelligence Research Assistant (AIRA) (2025) — link
  2. Detecting Suspicious Transactions Using Deep Learning-Based Process Mining (2025) — link
  3. Extracting Financial Data from Unstructured Sources: Leveraging Large Language Models (2024) — link
  4. Managers’ perception of product market competition and earnings management: a textual analysis of firms’ 10-K reports (2024) — link
  5. Changes in Risk Factor Disclosures and the Variance Risk Premium (2023) — link
  6. A Survey of Large Language Models (2023) — link
  7. Exploring the Feasibility of ChatGPT for Event Extraction (2023) — link
  8. FinBERT : A Large Language Model for Extracting Information from Financial Text† (2022) — link
and 36 more
  1. Textual Analysis in Accounting: What’s Next? (2022) — link
  2. What Makes Good In-Context Examples for GPT-3? (2021) — link
  3. Language Models are Few-Shot Learners (2020) — link
  4. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks (2019) — link
  5. Supervised and Unsupervised Neural Approaches to Text Readability (2019) — link
  6. Annual Report Readability, Tone Ambiguity, and the Cost of Borrowing (2017) — link
  7. Textual Analysis in Accounting and Finance: A Survey (2016) — link
  8. The benefits of specific risk-factor disclosures (2016) — link
  9. Bidirectional LSTM-CRF Models for Sequence Tagging (2015) — link
  10. Text and Context: Language Analytics in Finance (2014) — link
  11. Textual Sentiment in Finance: A Survey of Methods and Models (2014) — link
  12. Distributed Representations of Words and Phrases and their Compositionality (2013) — link
  13. The information content of mandatory risk factor disclosures in corporate filings (2013) — link
  14. Large-Sample Evidence on Firms’ Year-Over-Year MD&A Modifications (2011) — link
  15. Textual Analysis of Corporate Disclosures: A Survey of the Literature (2011)
  16. The Information Content of Forward-Looking Statements in Corporate Filings—A Naïve Bayesian Machine Learning Approach (2010) — link
  17. Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis (2010) — link
  18. The Effect of Disclosures by Management, Analysts, and Business Press on Cost of Capital, Return Volatility, and Analyst Forecasts: A Study Using Content Analysis (2009) — link
  19. Methodological Insights: Impression management: Developing and illustrating a scheme of analysis for narrative disclosures – a methodological note (2009) — link
  20. Design Challenges and Misconceptions in Named Entity Recognition (2009) — link
  21. Annual Report Readability, Current Earnings, and Earnings Persistence (2008) — link
  22. Information Asymmetry, Corporate Disclosure and the Capital Markets: A Review of the Empirical Disclosure Literature (2001) — link
  23. Voluntary Causal Disclosures: Tendencies and Capital Market Reaction (2000) — link
  24. Long Short-Term Memory (1997) — link
  25. Stock price crash risk and the managerial rhetoric mechanism: Evidence from R&D narrative disclosure in 10-K filings (2021)
  26. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (2019) — link
  27. What are you saying ? Using topic to detect financial misreporting * (2015)
  28. Lazy Prices * (2015) — link
  29. Forward-Looking MD&A Disclosures and the Information Environment (2015) — link
    1. Report Users’ Perceived Sentiments of Key Audit Matters and Firm Performance: Evidence from a Deep Learning-Based Natural Language Processing Approach
    1. The Future of Accounting: Sentiment Analysis in AI-Rewritten SEC Filings
    1. FiNER: Financial Numeric Entity Recognition for XBRL Tagging, May, at Dublin, Ireland
  33. Keywords: pre-trained language model, large language model, natural language processing, item segmentation, 10-K filing
    1. Large Language Models: A Survey
    1. GPT (Generative Pre-Trained Transformer)— A Comprehensive Review on Enabling Technologies, Potential Applications, Emerging Challenges, and Future Directions
    1. Large Language Model Is Not a Good Few-shot Information Extractor, but a Good Reranker for Hard Samples! Paper read at

0 of 44 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.