QuAnTS: Question Answering on Time Series
Paper • 2511.05124 • Published
Datasets:
sample_id int32 24k 27k | question_id int32 0 4 | trajectory array 3D | action_sequence dict | textual_description stringlengths 288 676 | question_type stringclasses 18
values | question stringlengths 29 151 | answer_type stringclasses 1
value | answer_text stringclasses 10
values | answer class label 2
classes |
|---|---|---|---|---|---|---|---|---|---|
24,000 | 3 | [[[-0.0,-0.0,0.9023759961128235],[-0.06914012879133224,-0.006030385848134756,0.810637891292572],[0.0(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["playing guitar","holding a baby","p(...TRUNCATED) | "The person is starting the sequence with playing guitar from 0:0.0 to 0:4.0 (for 4.0 seconds). Some(...TRUNCATED) | comparison_timestamp_same_binary | Is there no notable distinction between the actions at 15.608 and 7.648? | binary | No | 0true |
24,001 | 1 | [[[-0.0,-0.0,0.9063922166824341],[-0.07129761576652527,-0.003202479099854827,0.8162302374839783],[0.(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["kicking a ball","drinking with the (...TRUNCATED) | "The first activity the person is doing is kicking a ball from 0:0.0 to 0:4.0 (4.0 seconds total). T(...TRUNCATED) | first_binary | Was drinking with the left hand the first action performed? | binary | This is not correct. | 0true |
24,001 | 3 | [[[-0.0,-0.0,0.9063922166824341],[-0.07129761576652527,-0.003202479099854827,0.8162302374839783],[0.(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["kicking a ball","drinking with the (...TRUNCATED) | "The first activity the person is doing is kicking a ball from 0:0.0 to 0:4.0 (4.0 seconds total). T(...TRUNCATED) | comparison_timestamp_same_binary | Is there no notable distinction between the actions at 1.742 and 9.754? | binary | False | 0true |
24,001 | 4 | [[[-0.0,-0.0,0.9063922166824341],[-0.07129761576652527,-0.003202479099854827,0.8162302374839783],[0.(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["kicking a ball","drinking with the (...TRUNCATED) | "The first activity the person is doing is kicking a ball from 0:0.0 to 0:4.0 (4.0 seconds total). T(...TRUNCATED) | interval_part_sequence_binary | "Is it accurate to state that the person is engaged in exactly 2 distinct behaviors from 12.374 up t(...TRUNCATED) | binary | False | 0true |
24,002 | 0 | [[[0.0,-0.0,0.9123662114143372],[-0.07181760668754578,-0.006560855079442263,0.8226505517959595],[0.0(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["eating with the right hand","drinki(...TRUNCATED) | "The person initiates their actions with eating with the right hand from 0:0.0 to 0:4.0 (4.0 seconds(...TRUNCATED) | comparison_first_last_different_binary | Are the initial and final actions not the same? | binary | Not true | 0true |
24,002 | 1 | [[[0.0,-0.0,0.9123662114143372],[-0.07181760668754578,-0.006560855079442263,0.8226505517959595],[0.0(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["eating with the right hand","drinki(...TRUNCATED) | "The person initiates their actions with eating with the right hand from 0:0.0 to 0:4.0 (4.0 seconds(...TRUNCATED) | after_binary | Did eating with the right hand happen some time after drinking with the left hand for the person? | binary | This is correct! | 1false |
24,003 | 1 | [[[0.0,-0.0,0.9179534316062927],[-0.07290537655353546,-0.0018136876169592142,0.8291388750076294],[0.(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["picking something up with both hand(...TRUNCATED) | "The sequence starts with picking something up with both hands from 0:0.0 to 0:4.0 (for 4.0 seconds)(...TRUNCATED) | after_binary | "After the initial instance of picking something up with both hands, did the person later participat(...TRUNCATED) | binary | This is not correct. | 0true |
24,003 | 3 | [[[0.0,-0.0,0.9179534316062927],[-0.07290537655353546,-0.0018136876169592142,0.8291388750076294],[0.(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["picking something up with both hand(...TRUNCATED) | "The sequence starts with picking something up with both hands from 0:0.0 to 0:4.0 (for 4.0 seconds)(...TRUNCATED) | right_before_binary | A person was playing guitar. Were they dancing directly before that? | binary | No | 0true |
24,005 | 0 | [[[0.0,0.0,0.9131948351860046],[-0.07279270887374878,0.0005940566188655794,0.824286162853241],[0.064(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["shaking hands","punching","catching(...TRUNCATED) | "The sequence starts with shaking hands. This is happening from 0:0.0 to 0:4.0, which means for a to(...TRUNCATED) | comparison_timestamp_same_binary | Is the action at 8.188 no different from the action at 13.841? | binary | This is not correct. | 0true |
24,006 | 2 | [[[0.0,-0.0,0.9075813293457031],[-0.07509024441242218,-0.005268143489956856,0.8206049799919128],[0.0(...TRUNCATED) | {"start":[0.0,4.0,8.0,12.0],"end":[4.0,8.0,12.0,16.0],"action":["bowing","holding a baby","holding a(...TRUNCATED) | "The sequence starts with bowing from 0:0.0 to 0:4.0 (4.0 seconds total). Someone is holding a baby.(...TRUNCATED) | comparison_first_last_different_binary | Are the first and last actions unlike each other? | binary | This is correct! | 1false |
End of preview. Expand in Data Studio
QuAnTS: Question Answering on Time Series – Dataset
QuAnTS is a challenging dataset designed to bridge the gap in question-answering research on time series data. The dataset features a wide variety of questions and answers concerning human movements, presented as tracked skeleton trajectories. QuAnTS also includes human reference performance to benchmark the practical usability of models trained on this dataset.
At present, there is no official leaderboard for this dataset.
Dataset Generation Overview
For details, please refer to the paper.
Task and Format
The primary task for the QuAnTS dataset is Time Series Question Answering. Given a time series of human skeleton trajectories and a question in natural language, the goal is to generate a correct answer. Answers are provided in one of the following formats: binary (Yes/No), multiple-choice (A/B/C), or open (free text). Additionally, to provide more training data for free-text answers, we provide entirely textual answers for all binary and multiple-choice questions. The ground truth action sequence or scene descriptions may not be used to answer the dataset — we provide them for debugging purposes only. The text in the dataset is in English.
We provide fixed splits into training, validation, and test portions, where only the latter may be used to compare performance across different approaches. You are free to mix the training and validation splits as needed.
Licensing and Citation
The QuAnTS dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
If you use the QuAnTS dataset in your research, please cite the paper:
@article{divo2026quants,
title={QuAnTS: Question Answering on Time Series},
author={Divo, Felix and Kraus, Maurice and Nguyen, Anh Q. and Xue, Hao and Razzak, Imran and Salim, Flora D. and Kersting, Kristian and Dhami, Devendra Singh},
journal={Journal of Data-centric Machine Learning Research},
year={2026},
url={https://openreview.net/forum?id=bNCDElSOXB}
}
- Downloads last month
- 251