README.md

---
license: apache-2.0
task_categories:
  - text-classification
  - question-answering
language:
  - en
size_categories:
  - 100K<n<1M
---
# CuratedTextCorpus
<!-- markdownlint-disable first-line-h1 -->
<!-- markdownlint-disable html -->
<!-- markdownlint-disable no-duplicate-header -->

<div align="center">
  <img src="figures/fig1.png" width="60%" alt="CuratedTextCorpus" />
</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 CuratedTextCorpus dataset represents a major advancement in high-quality text data for NLP tasks. Through rigorous curation and validation processes, we have assembled a collection that meets the highest standards for machine learning applications. The dataset excels in text classification, question answering, and general language understanding tasks.

<p align="center">
  <img width="80%" src="figures/fig3.png">
</p>

Compared to previous versions, this curated dataset shows significant improvements in data quality metrics. For instance, in duplicate detection tests, the deduplication rate has improved from 85% to 99.2%. This advancement stems from our enhanced preprocessing pipeline that now includes semantic similarity checks in addition to exact matching.

Beyond improved deduplication, this version also offers reduced noise levels, better annotation consistency, and enhanced domain coverage.

## 2. Quality Metrics

### Comprehensive Quality Assessment

<div align="center">

| | Metric | Baseline | v1.0 | v2.0 | CuratedTextCorpus |
|---|---|---|---|---|---|
| **Data Completeness** | Completeness | 0.821 | 0.855 | 0.871 | 0.877 |
| | Consistency | 0.756 | 0.782 | 0.801 | 0.806 |
| | Accuracy | 0.689 | 0.721 | 0.745 | 0.751 |
| **Data Validity** | Validity | 0.812 | 0.834 | 0.856 | 0.861 |
| | Uniqueness | 0.901 | 0.925 | 0.941 | 0.945 |
| | Timeliness | 0.667 | 0.698 | 0.721 | 0.727 |
| **Data Integrity** | Integrity | 0.778 | 0.801 | 0.823 | 0.828 |
| | Relevance | 0.712 | 0.738 | 0.761 | 0.766 |
| | Coverage | 0.645 | 0.678 | 0.702 | 0.708 |
| **Additional Metrics** | Conformity | 0.834 | 0.856 | 0.878 | 0.883 |
| | Precision | 0.723 | 0.751 | 0.776 | 0.782 |
| | Reliability | 0.789 | 0.812 | 0.834 | 0.839 |

</div>

### Overall Quality Summary
The CuratedTextCorpus demonstrates exceptional quality across all evaluated metrics, with particularly strong results in completeness and integrity assessments.

## 3. Data Access & API
We provide direct access to the dataset through our data portal. Please check our official documentation for API access details.

## 4. How to Use

Please refer to our documentation for information on loading and using CuratedTextCorpus.

Usage recommendations for CuratedTextCorpus:

1. Preprocessing scripts are included for common NLP tasks.
2. Balanced sampling utilities are available for imbalanced labels.

The data format follows standard HuggingFace datasets conventions with train/validation/test splits.

### Loading the Dataset
```python
from datasets import load_dataset

dataset = load_dataset("username/CuratedTextCorpus")
```

### Data Fields
The dataset includes the following fields:
- `text`: The main text content
- `label`: Classification label (if applicable)
- `metadata`: Additional context information

### Recommended Preprocessing
```python
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

def preprocess(examples):
    return tokenizer(examples["text"], truncation=True, padding=True)

tokenized_dataset = dataset.map(preprocess, batched=True)
```

## 5. License
This dataset is licensed under the [Apache 2.0 License](LICENSE). The use of CuratedTextCorpus is subject to the license terms. Commercial use is permitted with attribution.

## 6. Contact
If you have any questions, please raise an issue on our repository or contact us at data@curatedtextcorpus.ai.