Datasets:

ccm
/

2025-24679-text-dataset

dataset_info:
  features:
    - name: haiku
      dtype: string
    - name: label
      dtype: int64
  splits:
    - name: original
      num_bytes: 5806
      num_examples: 72
    - name: augmented
      num_bytes: 23010
      num_examples: 288
  download_size: 16630
  dataset_size: 28816
configs:
  - config_name: default
    data_files:
      - split: original
        path: data/original-*
      - split: augmented
        path: data/augmented-*
license: mit
task_categories:
  - text-classification
language:
  - en
pretty_name: 24-679 Text Dataset

Dataset Card for `ccm/2025-24679-text-dataset`

This dataset contains short haiku poems paired with categorical labels. It was created as a class exercise for text classification tasks, with both original and augmented examples provided to enable experimentation with supervised learning workflows.

Dataset Details

Dataset Description

The dataset consists of haiku (short, three-line poems) paired with numeric labels for classification purposes. It was designed to be a small, approachable dataset for teaching text processing, feature extraction, and classification in natural language processing.

Curated by: Fall 2025 24-679 course at Carnegie Mellon University
Shared by [optional]: Christopher McComb
License: MIT
Language(s): English

Uses

Direct Use

Train text classification models (e.g., predicting label categories from haiku content).
Practice preprocessing, tokenization, and feature extraction in NLP pipelines.
Experiment with AutoML frameworks on text data.

Out-of-Scope Use

Generalization to real-world poetry classification tasks.
Cultural, literary, or stylistic analysis beyond the small synthetic dataset.
Any commercial or production-grade NLP system.

Dataset Structure

The dataset has two splits:

original: 72 examples of hand-written haiku with assigned labels.
augmented: 288 examples generated by augmentation to expand data diversity and balance.

Each row includes:

haiku (string): the text of the haiku.
label (int): a numeric class label.

Dataset Creation

Curation Rationale

The dataset was created to give students hands-on practice with NLP pipelines in a controlled, lightweight context. Haiku were chosen because they are short, interpretable, and easy to encode consistently.

Source Data

The haiku were either:

Written or adapted by students in the course.
Augmented using simple data generation and paraphrasing techniques.

Data Collection and Processing

Original haiku were composed during in-class activities.
Labels were assigned according to pre-defined categories.
Augmented examples were generated with rule-based or model-based text augmentation methods.

Who are the source data producers?

Original data: Students enrolled in the course.
Augmented data: Produced automatically by instructors and teaching assistants.

Bias, Risks, and Limitations

Small dataset: Only 72 original haiku.
Synthetic augmentation: Augmented haiku may be repetitive or less natural than real examples.
Cultural narrowness: Haiku were written by a limited, English-speaking student cohort.

Recommendations

Use primarily for teaching and demonstration purposes.
Do not draw literary or cultural conclusions from the dataset.
Emphasize limitations and bias awareness in classroom discussions.

Dataset Card Contact

Christopher McComb (Carnegie Mellon University) — [email protected]

Dataset Card for ccm/2025-24679-text-dataset