My News

A News app with Sentiment analysis, Recommendations and Data Visualization.

Overview

A React frontend for my backend service and REST API which polls popular RSS news feeds for articles and performs sentiment analysis on headlines and descriptions.

Main Features

• Classify news articles with their sentiment score from 1 to -1, positive/neutral/negative.
• User voting system for sentiment, inspired by hotukdeals and Reddit
• Articles filtering by metadata - sentiment rating, source and topic.
• Beautiful data visualization with MongoDB Atlas Charts.
• Recommendation system using OpenAI text embeddings - filter by semantically similar articles
  • Generates a joint text embedding of article titles, genre, description which is stored in a Pinecone vector database
  • Similar embeddings are queried and ranked based on the cosine distance between these vectors.

Technology Stack

• MongoDB Realm
• React
• Redux
• Mantine React Framework
• FastAPI
• Python transformers
• Pinecone - vector database

Backend

Article headlines and descriptions are given a positivity score between -1 and 1. The model used is based on BERT, via the transformers python package.

Sentiment results are stored with article metadata on a MongoDB database which is exposed to the web app.

For a more detailed technical breakdown and summary, visit Antony90/rss-news-sentiment-analysis.

Setup

Requirements

• Node.js
• Antony90/rss-news-sentiment-analysis to populate the database with articles.

Usage

1. Run the web server locally npm run start

2. Start the backend service to collect news articles.