Open Recommender - An open source AI-powered recommendation system for videos and articles

⚠️ Work in Progress... ⚠️

🚀 Overview

Welcome to Open Recommender, an open source AI-powered recommendation system for videos and articles.

🏆 Goals

• Understand your current interests by analyzing things like your Twitter engagement (likes, retweets, etc.) and other data sources.
• Search the web for interesting content to recommend to you
• Recommend interesting sections from videos
• Recommend "timeless" content rather than "trending" content
• Surface "niche bangers" - difficult to find but really high quality content
• Biased towards learning as opposed to entertainment
• Read my blog post for more: Building an LLM-Powered Open Source Recommendation System for YouTube
• Create a system that borrows the best elements from YouTube Shorts, TikTok and incremental reading to create something that feels as effortless and engaging as a queue of YouTube shorts but actually helps you make progress towards meaningful goals

How to Run

Installation

• git clone this repo
• cd open-recommender
• yarn && yarn build
• python3 -m venv env
• source env/bin/activate
• pip install -r requirements.txt
• In the client, server and cli packages, cp .env.example .env and fill in the values
• If you want to use Twitter as an input data source for recommendations, you need to create a fake Twitter account and create an accounts.txt file in the root folder with the account's credentials in the format username:password:email:email_password.

Running

• yarn server to run the backend
• yarn client to run the frontend
• yarn worker to run the background job worker
• Open up the web client and click the login button in the top right.
• After logging in using Twitter this will automatically trigger a new recommendations pipeline run task for the worker.
• You can monitor a recommendations pipeline run using by navigating to http://localhost:5173/#/admin. Make sure you set ADMIN="Your Twitter Username" in the server .env file.
• After a run is finished you can view your queue of recommendations by navigating to your feed.

Commands

📚 How it Works

A summary of the data pipeline:

• Download a user's Twitter data (tweets, likes, retweets, etc.)
• Recursively summarize into a user profile
• Generate search queries using the user's profile
• Search for videos and articles based using the queries
• Download transcripts and articles and chunk them into "clips"
• Recommend the best clips to the user in clusters (like mini learning playlists)

Papers and Blog Posts

• Building an LLM-Powered Open Source Recommendation System for YouTube
• BookGPT: A General Framework for Book Recommendation Based on a Large Language Model
• A Survey on Large Language Models for Recommendation
• Recommender Systems in the Era of Large Language Models (LLMs)