Model Card: Paraphrase Identification

Model Details

• Model Name: ParaBERT
• Description: A fine-tuned paraphrase identification model based on BERT
• Author: Lucie Gabagnou, Armand L'Huillier, Yanis Rehoune, Ghiles Idris
• Language: Pytorch

Intended Use

• Primary intended uses: This model is designed to identify whether two questions are paraphrases of each other.
• Primary intended users: This model is intended for use by NLP researchers and developers who are working on tasks related to paraphrase identification.
• Out-of-scope use cases: This model should not be used for tasks outside of paraphrase identification, or in situations where the input data may contain sensitive or confidential information.

Model Architecture and Training Data

• Model Architecture: BERT
• Training Data: https://huggingface.co/datasets/bigscience/P3/viewer/glue_qqp_same_thing/train (Only questions)

Evaluation Data and Results

• Evaluation Data: https://huggingface.co/datasets/bigscience/P3/viewer/glue_qqp_same_thing/test
• Metrics: Accuracy
• Results: 0.95