CMLA11: A Balanced Text Classification Benchmark of Long Articles

Official repository for CMLA11, a curated text classification dataset introduced in the EMNLP 2025 paper: "Context Minimization for Resource-Constrained Text Classification: Optimizing Performance-Efficiency Trade-offs through Linguistic Features" https://aclanthology.org/2025.findings-emnlp.833/.

Dataset Summary

CMLA11 is a custom, high-quality text classification benchmark containing 110,000 curated long articles across 11 diverse categories. The data is compiled from 26 distinct sources including major newspapers, blogs, and magazines, offering a balanced blend of American and British English texts.

With an average length of 716.64 tokens per article, CMLA11 is specifically designed to challenge models on complex, long-context classification, and to benchmark efficiency-oriented NLP frameworks like context minimization.

---

Dataset Structure

Data Splits

• train: 110,000 rows

Features

Each row in the dataset contains the following fields:

• text (string): The full text of the curated long article.
• category (string): The classification label (one of the 11 classes).

Class Labels (11 Categories)

The dataset covers a wide spectrum of topical domains:

1. arts
2. business
3. Healthcare
4. movies
5. nyregion
6. opinion
7. politics
8. sports
9. Tech
10. us
11. world

---

How to Use

You can easily load and stream this dataset directly using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("nahid-hub/CMLA11")

# View a single example
print(dataset['train'][0])

Citation

If you use CMLA11 in your research, please cite the official publication:

@inproceedings{hossain-kabir-2025-context,
    title = "Context Minimization for Resource-Constrained Text Classification: Optimizing Performance-Efficiency Trade-offs through Linguistic Features",
    author = "Hossain, Nahid  and
      Kabir, Md Faisal",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.findings-emnlp.833/",
    doi = "10.18653/v1/2025.findings-emnlp.833",
    pages = "15412--15426",
    ISBN = "979-8-89176-335-7"
}