michaelkrumdickkensho commited on
Commit
6e8f3ee
·
verified ·
1 Parent(s): 16c7e13

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +74 -32
README.md CHANGED
@@ -1,32 +1,74 @@
1
- ---
2
- license: apache-2.0
3
- configs:
4
- - config_name: default
5
- data_files:
6
- - split: train
7
- path: data/train-*
8
- dataset_info:
9
- features:
10
- - name: subject
11
- dtype: string
12
- - name: question
13
- dtype: string
14
- - name: A
15
- dtype: string
16
- - name: B
17
- dtype: string
18
- - name: C
19
- dtype: string
20
- - name: D
21
- dtype: string
22
- - name: answer
23
- dtype: string
24
- - name: task
25
- dtype: string
26
- splits:
27
- - name: train
28
- num_bytes: 52881
29
- num_examples: 131
30
- download_size: 36977
31
- dataset_size: 52881
32
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ configs:
4
+ - config_name: default
5
+ data_files:
6
+ - split: train
7
+ path: data/train-*
8
+ dataset_info:
9
+ features:
10
+ - name: subject
11
+ dtype: string
12
+ - name: question
13
+ dtype: string
14
+ - name: A
15
+ dtype: string
16
+ - name: B
17
+ dtype: string
18
+ - name: C
19
+ dtype: string
20
+ - name: D
21
+ dtype: string
22
+ - name: answer
23
+ dtype: string
24
+ - name: task
25
+ dtype: string
26
+ splits:
27
+ - name: train
28
+ num_bytes: 52881
29
+ num_examples: 131
30
+ download_size: 36977
31
+ dataset_size: 52881
32
+ ---
33
+
34
+ # Finance Fundamentals: Domain Knowledge
35
+
36
+ This dataset contains 131 multiple choice questions designed to test a models domain knowledge in business and finance. For more information, see the [BizBench paper.](https://aclanthology.org/2024.acl-long.452.pdf)
37
+
38
+
39
+ ## Example
40
+
41
+ ```
42
+ A fixed-rate system is characterized by:
43
+
44
+ A. explicit legislative commitment to maintain a specified parity.
45
+ B. monetary independence being subject to the maintenance of an exchange rate peg.
46
+ C. target foreign exchange reserves bearing a direct relationship to domestic monetary aggregates.
47
+ ```
48
+
49
+ ## Citation
50
+
51
+ If you find this data useful, please cite:
52
+ ```
53
+ @inproceedings{krumdick-etal-2024-bizbench,
54
+ title = "{B}iz{B}ench: A Quantitative Reasoning Benchmark for Business and Finance",
55
+ author = "Krumdick, Michael and
56
+ Koncel-Kedziorski, Rik and
57
+ Lai, Viet Dac and
58
+ Reddy, Varshini and
59
+ Lovering, Charles and
60
+ Tanner, Chris",
61
+ editor = "Ku, Lun-Wei and
62
+ Martins, Andre and
63
+ Srikumar, Vivek",
64
+ booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
65
+ month = aug,
66
+ year = "2024",
67
+ address = "Bangkok, Thailand",
68
+ publisher = "Association for Computational Linguistics",
69
+ url = "https://aclanthology.org/2024.acl-long.452/",
70
+ doi = "10.18653/v1/2024.acl-long.452",
71
+ pages = "8309--8332",
72
+ 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."
73
+ }
74
+ ```