Datasets:

kevinkyi
/

Homework1_text_dataset

pretty_name: Yelp-Style Short Reviews (Stars 1–5)
tags:
  - text
  - classification
  - augmentation
  - education
license: cc-by-4.0
task_categories:
  - text-classification

Yelp-Style Short Reviews (Stars 1–5)

Purpose. Educational dataset for practicing text data collection, light preprocessing, label design, augmentation, and publishing to the Hugging Face Hub for a university course.

Timestamp: 2025-09-15T22:22:36.082573Z Repo: datasets/kevinkyi/Homework1_text_dataset

Composition

Domain: short, user-authored review snippets (Yelp-like).
Samples: 100 originals, 1000 augmented.
Language: English.
Length: typically a few sentences.

Source / Collection

Original texts were manually collected (copy-typed) by the student from public Yelp pages for coursework.
Each row includes:
- text: the review snippet (free-form)
- label: integer star rating in [1, 2, 3, 4, 5]

Labels

label (target): star rating, integers 1–5
- 1 = very negative
- 5 = very positive

Class distribution (original split): {1: 64, 2: 15, 3: 6, 4: 6, 5: 9}

Preprocessing

Strip whitespace, ensure labels are ints in [1..5].
Removed empty rows.
No deduplication beyond manual checks.

Augmentation

Goal: reach ≥ 1,000 total synthetic examples without changing sentiment polarity.
Techniques (applied randomly per text):
- Synonym replacement (WordNet-based)
- Random deletion (low probability)
- Random swap (token-level)
- Character noise (small swaps/drops)

These are lightweight, label-preserving transformations. Augmented texts remain paired with the original star labels.

Splits

original: 100 rows
augmented: 1000 rows

Intended Use & Limits

Use: coursework, demos, experiments with text classification.
Limits: small sample size; domain- and style-specific; may include paraphrased content from external sources. Not suitable for production or for benchmarking beyond class use. No guarantee of balanced classes or real-world representativeness.

Ethical Notes

Contains no personal identifiers beyond publicly visible review text fragments.
When sharing beyond class, ensure texts comply with original platform terms and copyright.

License

CC BY 4.0 for the dataset curation and augmentation code. Text content should be used only in accordance with original site terms and for educational purposes.

AI Usage Disclosure

The majority of the notebook code was created by referencing the in class text dataset example. GenAI was used to assist with augmentation code, refactoring the in class example,and to help write helpful comments