SecureFinAI-Lab/Regulations_abbreviation

Overview

The dataset contains stock tickers and acronyms for regulatory terms. It is developed for the abbreviation recognition task at Regulations Challenge @ COLING 2025. This dataset is designed to evaluate and benchmark the performance of LLMs in understanding and generating expansions for abbreviations within the context of regulatory and compliance documentation. It provides a collection of abbreviations commonly encountered in regulatory texts, along with their full forms.

Legal and compliance documents, such as regulatory filings, compliance reports, and business contracts, contain a lot of abbreviations of legal terms, initialisms of financial agencies, and stock symbols. Misidentification of abbreviations may lead to misidentification of financial products and liable parties, resulting in misapplication of regulations, failure in compliance, and potentially serious legal and financial consequences. This task evaluates LLMs’ ability to give the full form of the stock symbols and the abbreviations in EMIR, SEC regulations, Federal Reserve regulations, and the accounting and auditing standards.

Statistics

Metrics

We measure the performance using the accuracy score that is calculated as the ratio of accurately identified abbreviations to the total number of abbreviations in the dataset.

License

The question set is licensed under CDLA-Permissive-2.0. It is a permissive open data license. It allows anyone to freely use, modify, and redistribute the dataset, including for commercial purposes, provided that the license text is included with any redistributed version. There are no restrictions on the use or licensing of any outputs, models, or results derived from the data.

Related tasks

Regulations Challenge at COLING 2025: https://coling2025regulations.thefin.ai/home