PDScience/CollegeAuditData

Dataset Description: College and University Financial Audits and Text Dataset

Overview

The College and University Financial Audits and Text Dataset is a dataset that combines audited financial statement texts with key extracted financial data from various colleges and universities. This dataset stands out for its emphasis on long context input data, making it especially useful for training large language models to extract data from long input texts.

Dataset Composition

This dataset contains over 125,000 entries with the following variables:

1. AuditText: The complete text of the audited financial statement. These can be very long input texts.
2. fiscalYear: The fiscal year of the reported data.
3. auditYear: The year in which the audit was performed.
4. value: Value of the variable in fiscal year==fiscalYear as reported in the audit text.
5. variable: Variable extracted from the audit text.

Key Features and Use Cases

• Long Context Input Data: The inclusion of complete audit texts offers a unique opportunity for training and testing LLMs on long-form financial documents.
• Model Training and fine-tuning: Ideal for fine-tuning AI models for data extraction from long textual inputs, such as in-depth financial reports.
• Financial and Educational Research: Facilitates comprehensive analysis in the domains of finance and education, allowing for nuanced insights into the financial workings of higher education institutions.

Suggested prompt format for training

Prompt: "Extract the {variable} for fiscal year {fiscalYear} from the following text: \n\n {AuditText}"

Response: "{value}"