|
|
--- |
|
|
license: apache-2.0 |
|
|
task_categories: |
|
|
- question-answering |
|
|
--- |
|
|
# SuperQA-Dataset |
|
|
<!-- markdownlint-disable first-line-h1 --> |
|
|
<!-- markdownlint-disable html --> |
|
|
<!-- markdownlint-disable no-duplicate-header --> |
|
|
|
|
|
<div align="center"> |
|
|
<img src="figures/fig1.png" width="60%" alt="SuperQA-Dataset" /> |
|
|
</div> |
|
|
<hr> |
|
|
|
|
|
<div align="center" style="line-height: 1;"> |
|
|
<a href="LICENSE" style="margin: 2px;"> |
|
|
<img alt="License" src="figures/fig2.png" style="display: inline-block; vertical-align: middle;"/> |
|
|
</a> |
|
|
</div> |
|
|
|
|
|
## 1. Introduction |
|
|
|
|
|
The SuperQA-Dataset represents a major advancement in question-answering benchmark datasets. In this latest release, we have significantly improved data quality through enhanced curation pipelines, rigorous validation processes, and comprehensive quality assurance measures. The dataset demonstrates exceptional performance across various data quality metrics, making it ideal for training and evaluating state-of-the-art QA models. |
|
|
|
|
|
<p align="center"> |
|
|
<img width="80%" src="figures/fig3.png"> |
|
|
</p> |
|
|
|
|
|
Compared to previous versions, the upgraded dataset shows significant improvements in data quality. For instance, in terms of completeness, the dataset has improved from 85% in the previous version to 96.5% in the current version. This advancement stems from enhanced data collection methodologies: our new pipeline processes 3x more source documents while maintaining stricter quality thresholds. |
|
|
|
|
|
Beyond improved completeness, this version also offers reduced noise, better answer coverage, and enhanced metadata annotations. |
|
|
|
|
|
## 2. Quality Assessment Results |
|
|
|
|
|
### Comprehensive Quality Metrics |
|
|
|
|
|
<div align="center"> |
|
|
|
|
|
| | Metric | Dataset-A | Dataset-B | Dataset-C | SuperQA-Dataset | |
|
|
|---|---|---|---|---|---| |
|
|
| **Core Quality Metrics** | Completeness | 0.845 | 0.867 | 0.882 | 0.965 | |
|
|
| | Accuracy | 0.891 | 0.903 | 0.912 | 0.972 | |
|
|
| | Consistency | 0.823 | 0.841 | 0.856 | 0.948 | |
|
|
| **Temporal & Validity** | Timeliness | 0.756 | 0.778 | 0.792 | 0.891 | |
|
|
| | Validity | 0.812 | 0.835 | 0.848 | 0.938 | |
|
|
| | Uniqueness | 0.934 | 0.942 | 0.951 | 0.987 | |
|
|
| | Integrity | 0.867 | 0.889 | 0.901 | 0.961 | |
|
|
| **Relevance & Access** | Relevance | 0.723 | 0.745 | 0.762 | 0.856 | |
|
|
| | Accessibility | 0.912 | 0.925 | 0.934 | 0.978 | |
|
|
| | Conformity | 0.834 | 0.851 | 0.867 | 0.945 | |
|
|
| | Precision | 0.878 | 0.895 | 0.908 | 0.968 | |
|
|
| **Advanced Quality**| Traceability | 0.689 | 0.712 | 0.731 | 0.823 | |
|
|
| | Representativeness | 0.756 | 0.778 | 0.795 | 0.889 | |
|
|
| | Portability | 0.823 | 0.845 | 0.861 | 0.941 | |
|
|
| | Credibility | 0.901 | 0.918 | 0.932 | 0.981 | |
|
|
|
|
|
</div> |
|
|
|
|
|
### Overall Quality Summary |
|
|
The SuperQA-Dataset demonstrates exceptional quality across all evaluated metric categories, with particularly notable results in accuracy, uniqueness, and credibility metrics. |
|
|
|
|
|
## 3. Dataset Explorer & API |
|
|
We offer a dataset explorer and API for you to interact with SuperQA-Dataset. Please check our official website for more details. |
|
|
|
|
|
## 4. How to Use |
|
|
|
|
|
Please refer to our code repository for more information about using SuperQA-Dataset. |
|
|
|
|
|
### Loading the Dataset |
|
|
```python |
|
|
from datasets import load_dataset |
|
|
|
|
|
dataset = load_dataset("your-username/SuperQA-Dataset-TestRepo") |
|
|
``` |
|
|
|
|
|
### Dataset Structure |
|
|
The dataset contains the following splits: |
|
|
- `train`: Training examples |
|
|
- `validation`: Validation examples |
|
|
- `test`: Test examples |
|
|
|
|
|
Each example contains: |
|
|
- `question`: The question text |
|
|
- `context`: Relevant context passage |
|
|
- `answer`: The correct answer |
|
|
- `metadata`: Additional annotations |
|
|
|
|
|
### Recommended Usage |
|
|
We recommend using this dataset with the following settings: |
|
|
1. Use stratified sampling for training |
|
|
2. Apply data augmentation carefully |
|
|
3. Validate model outputs against the provided answer formats |
|
|
|
|
|
### Data Format |
|
|
For loading, please follow the standard Hugging Face datasets format: |
|
|
```python |
|
|
from datasets import load_dataset |
|
|
|
|
|
# Load specific split |
|
|
train_data = load_dataset("your-username/SuperQA-Dataset-TestRepo", split="train") |
|
|
|
|
|
# Access examples |
|
|
for example in train_data: |
|
|
question = example["question"] |
|
|
answer = example["answer"] |
|
|
``` |
|
|
|
|
|
## 5. License |
|
|
This dataset is licensed under the [Apache 2.0 License](LICENSE). The use of SuperQA-Dataset is also subject to the [Apache 2.0 License](LICENSE). The dataset supports commercial use and derivative works. |
|
|
|
|
|
## 6. Contact |
|
|
If you have any questions, please raise an issue on our GitHub repository or contact us at data@superqa-dataset.ai. |
|
|
``` |
|
|
|
