DBQR-QA
A Question-Answering Dataset on a Hybrid of Database Querying and Reasoning
Submission starts 7 November 2024
A SHARED TASK
TLDR
Generate programs to answer a series of questions.
WHAT YOU GET
400 questions, ten per conversation, with tables (Pandas Data Frame) queried from our graph database.
YOUR MODEL GENERATES
Programs answering each question using our pre-defined (Python) or any custom functions in any language.
WHAT WE EVALUATE
The answer your model generated (number, text, set, or table).
Test the models' abilities to memorize and adapt to the changing logic throughout a conversation.
QUESTION 1
What is the ratio of combined revenues reported by companies in the retail industry with liabilities between 100M and 500M every year during 2017 and 2020 to those with liabilities between 300M and 1B during the same years?
QUESTION 4
If I removed the top three companies by combined deferred tax liabilities from the first group, what would be the ratio?
QUESTION 4
If I removed the top three companies by combined deferred tax liabilities from the first group, what would be the ratio?
Evaluation
Three-step automatic and manual evaluation.
AUTOMATIC
Evaluate using our Python evaluation script offline (unlimited) or online (up to 100 times per day).
GPT-4O
Close-to-human evaluation accuracy. Use our prompt (unlimited) or evaluate online (twice a week).
MANUAL
Manual evaluation at the end of the submission period with the release of test answers.
Dataset Statistics
Five subsets of questions by type and complexity.
100 ×
S
: Simple query with specific companies
View
100 ×
C
: Complex query with unspecified companies
View
50 ×
T
: Reasoning steps requiring multiple tables
View
100 ×
H
: Multiple-hop query
View
50 ×
I
: Instruction QA
View
Examples and practice data are now available.
See the quick-start page for more information.
Stages
Practice
50 questions
Training
200 questions
Blind test
150 questions
Workshop
@COLING 2025 in Abu Dhabi, UAE
19-20 January 2025
Important Dates
AoE time zone
Done
Practice set release 2 Sep 2024
Done
Training/test (blind) release 7 Nov 2024
In progress
First GPT evaluation 13 Nov 2024
Pending
Last GPT evaluation 23 Nov 2024
Pending
Paper submission deadline 25 Nov 2024
Pending
Test answers release 25 Nov 2024
Pending
Notification of acceptance 5 Dec 2024
Pending
Camera-ready deadline 13 Dec 2024
Participate
Register your email address and receive an access token to submit your answers.
Copyright © 2024-2025 R. Nararatwong et al. The images are generated by Dall-E · 3.
This website has been designed using resources from Flaticon.com. See credit page for links to materials used.
This study is partially based on the results obtained from a project JPNP20006,
commissioned by the New Energy and Industrial Technology Development Organization (NEDO).