metadata

license: apache-2.0
task_categories:
  - summarization
language:
  - en
size_categories:
  - 10K<n<100K
dataset_format: csv

EDGAR-CORPUS : 10K Financial Report Summarization

Extracted from SEC EDGAR filings (1993-2020). This dataset enhances financial report summarization by leveraging a hybrid AI model strategy.

Using:

ChatGPT-3.5 Turbo(~70%),

Claude 3.5 (~30% to generate structured, accurate, and concise summaries)

Dataset Composition

Summaries in this dataset are generated using a hybrid AI model strategy, balancing quality and efficiency:
ChatGPT-3.5 Turbo (~70%) – Used for structured, well-articulated summaries
Claude 3.5 (~30%) – Two variants applied based on input size:

Claude 3.5 Sonnet for longer and more detailed reports
Claude 3.5 Haiku for shorter reports to optimize processing speed

Enhanced Summarization with spaCy (Claude Exclusive)

The dataset incorporates spaCy for NLP-based word tracking, applied only when processing with Claude models. This ensures:

Balanced word distribution to prevent overfitting in AI training datasets
Improved entity recognition for financial terms like revenue, net income, liabilities
Adaptive model selection (Sonnet for long texts, Haiku for short ones)
Keyword-driven extraction of essential financial metrics (e.g., EBITDA, market cap, stock price)

Example Row from the Dataset

Below is a real example from the dataset, showing a financial statement input, the AI-generated summary, and the model used:

Claude's Summary Style

Model	Input (Financial Statement Text)	Summary (AI-Generated)
Claude	FACEBOOK, INC. INDEX TO CONSOLIDATED FINANCIAL STATEMENTS Reports of Independent Registered Public Accounting Firm Consolidated Financial Statements: Consolidated Balance Sheets... (Truncated for readability)	Here's a summary of the financial statement: Key Components: 1 Revenue Recognition - Revenue recognized when four conditions are met 2 Assets - Includes current assets, original programming costs, and amortization 3 Liabilities - Marketable securities, fair value measurement, and valuation hierarchy 4 Income Taxes - Deferred tax assets, valuation allowances, and tax rate impact Notable Accounting Practices: - Conservative approach to revenue recognition - Detailed fair value measurement hierarchy - Comprehensive tax planning and recognition strategy - Regular review of asset values and potential impairment

ChatGPT's Summary style:

Model	Input (Financial Statement Text)	Summary (AI-Generated)
ChatGPT	To the Shareholders and Board of Directors of Exactus, Inc. Opinion on the Financial Statements… (Truncated)	The financial statement provided indicates that the company has incurred losses from its operations, resulting in negative cash flows from operating activities. Additionally, the company has an accumulated deficit, and there is substantial doubt about its ability to continue as a going concern. This means that there are concerns about the company's ability to meet its financial obligations and continue operating in the foreseeable future. In terms of expenses, the company has liabilities that are part of its normal course of business operations. Despite the uncertainties surrounding the company's ability to continue as a going concern, no adjustments have been made to the carrying amount and classification of the company's assets and liabilities. The company has considered ASU 2014, which likely pertains to accounting standards related to going concern issues. The liabilities mentioned in the financial statement primarily consist of debt. It is essential to note that the company's financial position seems to be precarious, given the accumulated losses, negative cash flows, and doubts about its ability to continue operating. The management's evaluation of the situation, as described in Note 2 of the financial statement, likely outlines the events and conditions that have led to this financial position and the management's plans to address these challenges.

To Load this Dataset

from datasets import load_dataset

dataset = load_dataset("kritsadaK/EDGAR-CORPUS-Financial-Summarization")

Limitations: Summary Accuracy & AI Hallucinations

AI-generated financial summaries are not 100% accurate without safeguards. This dataset incorporates:

Entity Matching & Verification: spaCy NER cross-checks extracted financial terms.
Consistency Checks: Rule-based validation prevents misinterpretation (e.g., net loss ≠ profit trend).
Standardized Format: Aligns Claude and ChatGPT outputs for consistency.
Bias & Overfitting Prevention: Balances structured data extraction with AI-generated summaries.

Despite these safeguards, users should verify summaries before relying on them for financial decisions.

The Financial Statements Summary 10K Dataset was developed as part of the CSX4210: Natural Language Processing project at Assumption University.