Twitter Sentiment Dashboard

This repository contains an NLP sentiment-analysis project based on a BERT fine-tuning workflow. The notebook trains a transformer sequence classifier for positive, negative, and neutral social-media sentiment.

The hosted Streamlit app is a portfolio-ready interactive dashboard. It lets visitors:

• analyze a single tweet/review-style text
• compare multiple texts in a batch view
• inspect positive and negative word signals
• understand the original BERT training workflow
• see what model artifacts are required for real BERT inference online

Streamlit App

Open the live Streamlit app

Run locally:

pip install -r requirements.txt
python -m streamlit run app.py

For Streamlit Cloud:

Main file path: app.py

Important Model Note

The original app loaded a fine-tuned BERT model from a local machine path:

models/sent_model

Those trained model/tokenizer artifacts are not included in this repository, so the hosted dashboard uses a transparent lexical analyzer instead of pretending to run unavailable BERT weights.

The original BERT app code is preserved as:

legacy_bert_app.py

Project Files

.
|-- app.py                    # Streamlit Cloud dashboard
|-- legacy_bert_app.py        # Original BERT inference app
|-- fine-tune-bert-model.ipynb # BERT fine-tuning notebook
|-- requirements.txt          # Lightweight app dependencies
`-- README.md

BERT Workflow

The notebook covers:

1. Loading a Twitter sentiment dataset.
2. Cleaning and preparing tweet text.
3. Encoding positive, negative, and neutral labels.
4. Tokenizing text with bert-base-uncased.
5. Fine-tuning TFBertForSequenceClassification.
6. Evaluating the model on a test split.
7. Saving tokenizer/model artifacts for deployment.

Next Improvements

• Add the trained BERT tokenizer/model files or a reliable hosted model path.
• Replace the lexical analyzer with cached Hugging Face inference.
• Add test-set metrics directly to the Streamlit dashboard.
• Add confusion matrix and per-class precision/recall/F1.
• Add example social-media monitoring use cases.