license: cc0-1.0
task_categories:
- summarization
- text-retrieval
language:
- en
tags:
- legal
- law
size_categories:
- n<1K
source_datasets:
- FiscalNote/billsum
dataset_info:
- config_name: default
features:
- name: query-id
dtype: string
- name: corpus-id
dtype: string
- name: score
dtype: float64
splits:
- name: test
num_examples: 500
- config_name: corpus
features:
- name: _id
dtype: string
- name: title
dtype: string
- name: text
dtype: string
splits:
- name: corpus
num_examples: 500
- config_name: queries
features:
- name: _id
dtype: string
- name: text
dtype: string
splits:
- name: queries
num_examples: 500
configs:
- config_name: default
data_files:
- split: test
path: data/default.jsonl
- config_name: corpus
data_files:
- split: corpus
path: data/corpus.jsonl
- config_name: queries
data_files:
- split: queries
path: data/queries.jsonl
pretty_name: BillSumCA MTEB Benchmark
BillSumCA MTEB Benchmark 🏋
This is the Californian test split of the BillSum dataset formatted in the Massive Text Embedding Benchmark (MTEB) information retrieval dataset format.
This dataset is intended to facilitate the consistent and reproducible evaluation of information retrieval models on BillSum with the mteb embedding model evaluation framework.
More specifically, this dataset tests the ability of information retrieval models to retrieve Californian bills based on their summaries.
This dataset has been processed into the MTEB format by Isaacus, a legal AI research company.
Methodology 🧪
To understand how BillSum itself was created, refer to its documentation.
This dataset was formatted by taking the Californian split of BillSum, treating summaries as queries (or anchors) and bills as relevant (or positive) passages, and randomly sampling 500 examples (as per MTEB guidelines, to keep the size of this evaluation set manageable).
Structure 🗂️
As per the MTEB information retrieval dataset format, this dataset comprises three splits, default, corpus and queries.
The default split pairs summary (query-id) with the raw text of the bills (corpus-id), each pair having a score of 1.
The corpus split contains bills, with the text of a bill being stored in the text key and its id being stored in the _id key.
The queries split contains summaries, with the text of a summary being stored in the text key and its id being stored in the _id key.
License 📜
To the extent that any intellectual property rights reside in the contributions made by Isaacus in formatting and processing this dataset, Isaacus licenses those contributions under the same license terms as the source dataset. You are free to use this dataset without citing Isaacus.
The source dataset is licensed under CC0.
Citation 🔖
@inproceedings{Eidelman_2019,
title={BillSum: A Corpus for Automatic Summarization of US Legislation},
url={http://dx.doi.org/10.18653/v1/D19-5406},
DOI={10.18653/v1/d19-5406},
booktitle={Proceedings of the 2nd Workshop on New Frontiers in Summarization},
publisher={Association for Computational Linguistics},
author={Eidelman, Vladimir},
year={2019},
pages={48–56},
eprint={1910.00523}
}