README.md

metadata

tags:
  - text-classification
  - data-augmentation
  - superhero-texts

Superhero Texts Dataset

Dataset Description

Purpose

This dataset was created to explore text classification on descriptions of superheroes from different comic book universes (DC and Marvel). The original data consists of self-authored texts, and the augmented data expands this with synthetically generated variants using data augmentation techniques to increase the dataset size and diversity for potential model training.

Composition

The dataset contains descriptions of various superheroes and is labeled with the comic universe they belong to, either 'DC' or 'Marvel'.

Collection

The original data was self-authored. The augmented data was generated programmatically from the original data using data augmentation techniques.

Preprocessing

The original data was loaded from a CSV file. Basic data cleaning like handling missing values and duplicates was considered during the process, but the primary focus was on generating augmented data from the existing clean original data.

Labels

The dataset has a binary target label indicating the comic universe:

• DC
• Marvel

Splits

The dataset is divided into two splits:

• original: Contains the 100 self-authored texts.
• augmented: Contains 1000 augmented texts generated from the original data using EDA and character-level noise techniques.

Intended Use

This dataset is intended for training and evaluating text classification models to distinguish between descriptions of DC and Marvel superheroes. It can be used for educational purposes, research on data augmentation techniques for text classification, and as a starting point for building more complex models.

Limitations

• The dataset is relatively small, even with augmentation, and may not be representative of all possible ways to describe superheroes.
• The augmentation techniques used (EDA and character-level noise) are relatively simple and may not capture complex linguistic variations.
• The self-authored nature of the original data may introduce personal biases or specific writing styles.
• The model trained on this dataset might not generalize well to descriptions of superheroes from other universes or to different styles of text.

Ethical Considerations

The dataset contains fictional content about superheroes and does not directly involve sensitive personal information. However, when used in machine learning applications, it's important to be aware of potential biases that could arise from the original data or the augmentation process. Users should evaluate model performance carefully across different sub-groups if applicable and consider the potential for unintended consequences in downstream applications.

License

The dataset is released under the Apache-2.0 License.

AI Usage Disclosure

This dataset includes synthetically generated data in the 'augmented' split. The augmented data was created using Python code implementing Easy Data Augmentation (EDA) techniques (synonym replacement, random deletion, random swap, random insertion) and character-level noise (dropping and swapping characters). These techniques were applied to the original self-authored texts to increase the dataset size for model training purposes.