| --- |
| dataset_info: |
| features: |
| - name: question |
| dtype: string |
| - name: answer |
| dtype: float64 |
| - name: context |
| dtype: string |
| - name: task |
| dtype: string |
| splits: |
| - name: train |
| num_bytes: 638720 |
| num_examples: 223 |
| download_size: 198425 |
| dataset_size: 638720 |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train-* |
| --- |
| |
|
|
| # Finance Fundamentals: Quantity Extraction |
|
|
| This dataset contains evaluations for extracting numbers from financial text. The source data comes from: |
| - [TatQA](https://arxiv.org/abs/2105.07624) |
| - [ConvFinQA](https://arxiv.org/abs/2210.03849) |
|
|
| Each question went through additional manual review to ensure both correctness and clarity. For more information, see the [BizBench paper.](https://aclanthology.org/2024.acl-long.452.pdf) |
|
|
| ## Example |
| Each question will contain a document context: |
| ``` |
| The Company’s top ten clients accounted for 42.2%, 44.2% and 46.9% of its consolidated revenues during the years ended December 31, 2019, 2018 and 2017, respectively. |
| The following table represents a disaggregation of revenue from contracts with customers by delivery location (in thousands): |
| | | | Years Ended December 31, | | |
| | :--- | :--- | :--- | :--- | |
| | | 2019 | 2018 | 2017 | |
| | Americas: | | | | |
| | United States | $614,493 | $668,580 | $644,870 | |
| | The Philippines | 250,888 | 231,966 | 241,211 | |
| | Costa Rica | 127,078 | 127,963 | 132,542 | |
| | Canada | 99,037 | 102,353 | 112,367 | |
| | El Salvador | 81,195 | 81,156 | 75,800 | |
| | Other | 123,969 | 118,620 | 118,853 | |
| | Total Americas | 1,296,660 | 1,330,638 | 1,325,643 | |
| | EMEA: | | | | |
| | Germany | 94,166 | 91,703 | 81,634 | |
| | Other | 223,847 | 203,251 | 178,649 | |
| | Total EMEA | 318,013 | 294,954 | 260,283 | |
| | Total Other | 89 | 95 | 82 | |
| | | $1,614,762 | $1,625,687 | $1,586,008 | |
| ``` |
|
|
| An associated question that references the context: |
| ``` |
| What was the Total Americas amount in 2019? (thousand) |
| ``` |
|
|
| And an answer represented as a single float value: |
| ``` |
| 1296660.0 |
| ``` |
|
|
| ## Citation |
|
|
| If you find this data useful, please cite: |
| ``` |
| @inproceedings{krumdick-etal-2024-bizbench, |
| title = "{B}iz{B}ench: A Quantitative Reasoning Benchmark for Business and Finance", |
| author = "Krumdick, Michael and |
| Koncel-Kedziorski, Rik and |
| Lai, Viet Dac and |
| Reddy, Varshini and |
| Lovering, Charles and |
| Tanner, Chris", |
| editor = "Ku, Lun-Wei and |
| Martins, Andre and |
| Srikumar, Vivek", |
| booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", |
| month = aug, |
| year = "2024", |
| address = "Bangkok, Thailand", |
| publisher = "Association for Computational Linguistics", |
| url = "https://aclanthology.org/2024.acl-long.452/", |
| doi = "10.18653/v1/2024.acl-long.452", |
| pages = "8309--8332", |
| abstract = "Answering questions within business and finance requires reasoning, precision, and a wide-breadth of technical knowledge. Together, these requirements make this domain difficult for large language models (LLMs). We introduce BizBench, a benchmark for evaluating models' ability to reason about realistic financial problems. BizBench comprises eight quantitative reasoning tasks, focusing on question-answering (QA) over financial data via program synthesis. We include three financially-themed code-generation tasks from newly collected and augmented QA data. Additionally, we isolate the reasoning capabilities required for financial QA: reading comprehension of financial text and tables for extracting intermediate values, and understanding financial concepts and formulas needed to calculate complex solutions. Collectively, these tasks evaluate a model{'}s financial background knowledge, ability to parse financial documents, and capacity to solve problems with code. We conduct an in-depth evaluation of open-source and commercial LLMs, comparing and contrasting the behavior of code-focused and language-focused models. We demonstrate that the current bottleneck in performance is due to LLMs' limited business and financial understanding, highlighting the value of a challenging benchmark for quantitative reasoning within this domain." |
| } |
| ``` |
