Natural Language Processing (NLP) Experiments

Welcome to the NLP Experiments repository! This repository contains Jupyter notebooks demonstrating fundamental NLP techniques, including word embeddings, summarization, named entity recognition (NER), feature engineering, and word sense disambiguation (WSD). Each experiment provides hands-on implementations, evaluations, and best practices for working with text data.

📌 Repository Overview

This repository covers a range of NLP tasks using classical and modern techniques. The experiments leverage frameworks such as Python, TensorFlow, PyTorch, HuggingFace Transformers, and NLP libraries like spaCy and NLTK.

🏆 Experiments & Notebooks

1️⃣ GloVe: Implementing Word Embeddings from Scratch

• Task: Implement GloVe embeddings from scratch in Python.
• Key Topics: Word embeddings, co-occurrence matrices, vector representation.
• Dataset: Brown dataset (first 350 sentences) and WordNet synonyms.
• Evaluation: Mean Absolute Error (MAE) on cosine similarity between synonyms.
• Notebook: [Link]

2️⃣ Summarization with a Language Model

• Task: Train a language model to generate text summaries.
• Key Topics: Extractive & abstractive summarization, Transformer-based models.
• Evaluation: Rouge-L score between generated and actual summaries.
• Notebook: [Link]

3️⃣ Named Entity Recognition (NER) for Recipes

• Task: Apply NER techniques to identify ingredients in recipes.
• Key Topics: Sequence labeling, spaCy & Transformer-based NER models.
• Evaluation: Macro F1 score.
• Notebook: [Link]

4️⃣ Feature Engineering Using Parsing for Text Classification

• Task: Classify medical transcriptions into medical specialties using constituency & dependency parsing.
• Key Topics: Feature extraction, syntactic parsing, NLP classification.
• Evaluation: Macro F1 score.
• Notebook: [Link]

5️⃣ Word Sense Disambiguation (WSD) with Lesk Algorithm

• Task: Solve WSD by implementing the Lesk algorithm.
• Key Topics: Sense disambiguation, WordNet dictionary, text similarity.
• Evaluation: Levenshtein distance (Lower is better) between predicted and ground truth definitions from Senseval dataset.
• Notebook: [Link]

---

🔧 Tools & Frameworks Used

• Python 🐍
• TensorFlow/Keras 🏗️
• PyTorch 🔥
• HuggingFace Transformers 🤗
• Scikit-learn 📊
• NLTK & spaCy 📖
• Matplotlib & Seaborn 🎨

🤝 Contributing

Contributions are welcome! Feel free to open issues, suggest improvements, or add new NLP-related notebooks.

📜 License

This repository is licensed under the MIT License - see the LICENSE file for details.

---

🚀 Happy experimenting with NLP!