README.md

metadata

license: mit
task_categories:
  - text-classification
language:
  - en
size_categories:
  - 1K<n<10K

Paraphrase Detection Dataset (Derived from SetFit/stsb)

Description:

This dataset originates from the SetFit/stsb dataset, which was initially created for semantic textual similarity (STS) tasks with a label range of 0 to 5. It has been adapted for binary paraphrase detection by leveraging the high-accuracy paraphrase classification model viswadarshan06/pd-robert.

Each sentence pair in the original dataset has been re-labeled according to the following binary scheme:

• 1 → Paraphrase: The two sentences convey the same meaning.
• 0 → Not Paraphrase: The two sentences have different meanings.

This binary labeling makes the dataset directly applicable for paraphrase detection tasks within Natural Language Processing (NLP). It is particularly useful for:

• Training paraphrase detection models.
• Evaluating the performance of paraphrase detection models.
• Facilitating transfer learning for binary classification tasks related to sentence similarity.

Dataset Features:

The dataset contains the following features for each instance:

• sentence1: The first sentence in English.
• sentence2: The second sentence in English.
• label: A binary label indicating whether the two sentences are paraphrases:
  • 1: The sentences are paraphrases.
  • 0: The sentences are not paraphrases.

Use Cases:

This dataset is suitable for a variety of NLP applications, including:

• Paraphrase detection model training: Training new models to accurately identify paraphrases.
• Sentence similarity tasks: Evaluating how well models can determine if two sentences have similar meanings.
• Fine-tuning binary classification models: Adapting pre-trained binary classification models for the specific task of paraphrase detection.