Overview

This repository contains several NLP projects and experiments, including:

• Spam classification using large language models (LLMs)
• Sentiment analysis on tweets
• Topic modeling on the 20 Newsgroups dataset
• A Streamlit chatbot interface for local LLM inference

Contents

Notebooks & Scripts

• LLM_Usage_part_1.ipynb: Classifies emails as spam or ham using a Hugging Face LLM (zero-shot classification) on the spam.csv dataset. Evaluates accuracy, F1, recall, and confusion matrix.
• LLM_Usage_part_2.py: Streamlit app for a local chatbot using the Flan-T5-small model. Allows interactive Q&A with an LLM running locally.
• Sentiment_analysis_part_1.ipynb: Performs sentiment analysis on tweets using the VADER sentiment analyzer. Includes text cleaning and labeling.
• Sentiment_analysis_part_2.ipynb: Performs sentiment analysis on tweets using a Hugging Face transformer pipeline. Includes text cleaning and batch inference.
• Topic_modeling.ipynb: Topic modeling on the newsgroups dataset using LDA and NMF. Visualizes topics with bar charts and word clouds.

Datasets

• spam.csv: SMS messages labeled as spam or ham (not spam). Used for spam classification.
• tweets-data.csv: Tweets with metadata (date, likes, hashtags, etc.). Used for sentiment analysis.
• newsgroups: Binary (pickled) file containing the 20 Newsgroups dataset or similar. Used for topic modeling.

Setup & Requirements

Install the required Python packages:

pip install transformers pandas tqdm nltk vaderSentiment scikit-learn matplotlib wordcloud streamlit torch

For notebooks using NLTK, you may need to download resources:

import nltk
nltk.download('stopwords')
nltk.download('punkt')

Usage

• Open the Jupyter notebooks (.ipynb) for step-by-step code and explanations.
• Run LLM_Usage_part_2.py with Streamlit:streamlit run LLM_Usage_part_2.py
• Place the datasets (spam.csv, tweets-data.csv, newsgroups) in the project root.

Notes

• The newsgroups file is binary and should be loaded with Python’s pickle module.
• Some scripts may require GPU or MPS support for faster inference.

---

Feel free to adapt or extend these notebooks for your own NLP experiments!

The repo is accessible HERE