NewsSentiment

A lightweight Streamlit app for live RSS headline sentiment monitoring.

It fetches headlines from public RSS feeds, scores them using a fast rule-based economic/news sentiment classifier, and shows live dashboard metrics, confidence labels, and downloadable CSV results. Features

Live RSS feed monitoring
Fast headline sentiment classification
Confidence scoring
Positive / Negative / Neutral counts
CSV export of current snapshot
Optional auto refresh
Streamlit-friendly deployment

Project structure

NewsSentiment/ ├── app/ │ └── rss_sentiment_dashboard.py ├── tests/ │ └── test_classifier.py ├── requirements.txt ├── .gitignore └── README.md

Run locally

python -m venv .venv source .venv/bin/activate pip install -r requirements.txt streamlit run app/rss_sentiment_dashboard.py

Deploy from GitHub Streamlit Community Cloud

Push this repository to GitHub.
Go to Streamlit Community Cloud.
Create a new app.
Select your GitHub repo and branch.
Set the main file path to:

app/rss_sentiment_dashboard.py

Click Deploy.

Notes

This project uses a fast heuristic classifier instead of a large language model for live refresh performance. That makes it much faster and cheaper to run, but it can still misclassify ambiguous headlines. Suggested next improvements

Add manual review mode for low-confidence headlines
Add benchmark dataset and evaluation metrics
Add hybrid LLM fallback for uncertain cases
Add screenshot/GIF to this README

Lightweight LLM option

This app can optionally use a lightweight Hugging Face transformer fallback for low-confidence headlines. It uses cardiffnlp/twitter-roberta-base-sentiment-latest, which is much smaller and easier to deploy than a large generative model while still improving some ambiguous classifications [web:67].

To enable it locally, install the extra dependencies from requirements.txt and tick Enable lightweight LLM fallback in the sidebar. run with token

export HF_TOKEN="hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" streamlit run app/rss_sentiment_dashboard.py