twitter/the-algorithm

Features

• Content Discovery Algorithms - Generate candidate Tweets from outside a user's network by traversing engagement graphs to identify relevant content and similar user interests.
• Content Ranking Models - Predict the relevance of candidate Tweets using a neural network trained on interaction data to score and rank content for the timeline.
• Social Feed Ranking Algorithms - Retrieve relevant Tweets from a user's network by ranking them with a logistic regression model based on engagement likelihood between users.
• Timeline Construction Services - Construct and serve personalized content timelines by coordinating candidate sourcing, ranking models, and visibility filtering services to process user interaction data.
• Candidate Sourcing Pipelines - A multi-stage retrieval architecture that aggregates content from diverse social graphs and embedding spaces before final ranking.
• Neural Ranking Models - A deep learning scoring system that predicts user engagement probabilities by processing interaction features through multi-layer neural architectures.
• Recommendation Engine Pipelines - A distributed architecture that orchestrates candidate sourcing, neural network ranking, and heuristic filtering to deliver personalized content feeds.
• Feed Composition Engines - Blend ranked Tweets with non-Tweet content like ads and recommendations to finalize the timeline display for the user.
• Personalized Feed Orchestrators - A multi-stage content assembly layer that blends diverse media types and social signals into a unified, ranked user experience.
• Similarity Search Engines - A vector-based retrieval mechanism that identifies relevant content by calculating geometric proximity between user and tweet representations in high-dimensional space.
• Model Serving Environments - A high-performance execution environment that deploys predictive models to score content relevance and user engagement probabilities in real-time.
• Multi-Task Learning Models - A shared model architecture that predicts multiple engagement signals simultaneously to optimize content relevance across different platform surfaces.
• Embedding-Based Retrieval - Calculate similarity between users and content using numerical representations to identify relevant Tweets outside of a user's immediate social network.
• Graph-Based Content Discovery - A data processing architecture that traverses social and interest-based connections to identify relevant content outside of a user's immediate network.
• Feed Filtering Heuristics - Apply heuristics and filters to the ranked feed to ensure content diversity, visibility preferences, and social quality safeguards.
• Content Filtering Heuristics - A post-ranking processing layer that applies safety, diversity, and visibility constraints to ensure content quality before final delivery.
• Graph Traversal Strategies - A data-retrieval strategy that traverses social and interaction edges to discover relevant content outside of a user's immediate network.
• ML Serving Infrastructure - Manage shared data services, machine learning models, and high-performance serving frameworks that power recommendation and interaction features across the platform.
• Personalized Notification Engines - Surface personalized content recommendations via push notifications using multi-task learning models to predict user engagement probabilities and relevance.