Sentence-Embeddings-Using-N-Gram-Features-and-Contrastive-Learning-for-Multilingual-Datasets

This project explores a novel approach to generating sentence embeddings by leveraging N-gram features combined with contrastive learning. It aims to improve semantic representation across multilingual datasets, especially for low-resource languages.

---

🚀 Features

• Sentence embeddings using character and word N-grams
• Contrastive learning framework to align similar sentences
• Works across multilingual and domain-diverse datasets
• Evaluation using cosine similarity and clustering visualization

---

📁 Datasets Used

• Harry Potter and The Half-Blood Prince.txt – Fictional narrative
• bible.txt – Multilingual structured text
• Dataset25.txt – Custom multilingual dataset

---

🛠️ Setup & Usage

1. Clone the repository

   git clone https://github.com/MS134340/Sentence-Embeddings-Using-N-Gram-Features-and-Contrastive-Learning-for-Multilingual-Datasets.git
   cd Sentence-Embeddings-Using-N-Gram-Features-and-Contrastive-Learning-for-Multilingual-Datasets
   

2. Install required libraries

   pip install numpy pandas scikit-learn matplotlib seaborn
   

3. Run the notebook Open and execute the .ipynb file in Jupyter or Google Colab.

📊 Results Summary

• Outperformed simple TF-IDF and averaging methods.
• Embeddings better captured sentence-level semantic similarity.
• Contrastive loss improved clustering of similar meanings.

🧠 Techniques Used

• N-gram Feature Engineering
• Sentence Vector Aggregation
• Supervised Contrastive Loss
• Cosine Similarity Evaluation