the-algorithm README

X's Recommendation Algorithm

X's Recommendation Algorithm is a set of services and jobs that are responsible for serving feeds of posts and other content across all X product surfaces (e.g. For You Timeline, Search, Explore, Notifications). For an introduction to how the algorithm works, please refer to our engineering blog.

Architecture

Product surfaces at X are built on a shared set of data, models, and software frameworks. The shared components included in this repository are listed below:

Type	Component	Description
Data	tweetypie	Core service that handles the reading and writing of post data.
	unified-user-actions	Real-time stream of user actions on X.
	user-signal-service	Centralized platform to retrieve explicit (e.g. likes, replies) and implicit (e.g. profile visits, tweet clicks) user signals.
Model	SimClusters	Community detection and sparse embeddings into those communities.
	TwHIN	Dense knowledge graph embeddings for Users and Posts.
	trust-and-safety-models	Models for detecting NSFW or abusive content.
	real-graph	Model to predict the likelihood of an X User interacting with another User.
	tweepcred	Page-Rank algorithm for calculating X User reputation.
	recos-injector	Streaming event processor for building input streams for GraphJet based services.
	graph-feature-service	Serves graph features for a directed pair of users (e.g. how many of User A's following liked posts from User B).
	topic-social-proof	Identifies topics related to individual posts.
	representation-scorer	Compute scores between pairs of entities (Users, Posts, etc.) using embedding similarity.
Software framework	navi	High performance, machine learning model serving written in Rust.
	product-mixer	Software framework for building feeds of content.
	timelines-aggregation-framework	Framework for generating aggregate features in batch or real time.
	representation-manager	Service to retrieve embeddings (i.e. SimClusers and TwHIN).
	twml	Legacy machine learning framework built on TensorFlow v1.

The product surfaces currently included in this repository are the For You Timeline and Recommended Notifications.

For You Timeline

The diagram below illustrates how major services and jobs interconnect to construct a For You Timeline.

The core components of the For You Timeline included in this repository are listed below:

Type	Component	Description
Candidate Source	search-index	Find and rank In-Network posts. ~50% of posts come from this candidate source.
	tweet-mixer	Coordination layer for fetching Out-of-Network tweet candidates from underlying compute services.
	user-tweet-entity-graph (UTEG)	Maintains an in memory User to Post interaction graph, and finds candidates based on traversals of this graph. This is built on the GraphJet framework. Several other GraphJet based features and candidate sources are located here.
	follow-recommendation-service (FRS)	Provides Users with recommendations for accounts to follow, and posts from those accounts.
Ranking	light-ranker	Light Ranker model used by search index (Earlybird) to rank posts.
	heavy-ranker	Neural network for ranking candidate posts. One of the main signals used to select timeline posts post candidate sourcing.
Post mixing & filtering	home-mixer	Main service used to construct and serve the Home Timeline. Built on product-mixer.
	visibility-filters	Responsible for filtering X content to support legal compliance, improve product quality, increase user trust, protect revenue through the use of hard-filtering, visible product treatments, and coarse-grained downranking.
	timelineranker	Legacy service which provides relevance-scored posts from the Earlybird Search Index and UTEG service.

Recommended Notifications

The core components of Recommended Notifications included in this repository are listed below:

Type	Component	Description
Service	pushservice	Main recommendation service at X used to surface recommendations to our users via notifications.
Ranking	pushservice-light-ranker	Light Ranker model used by pushservice to rank posts. Bridges candidate generation and heavy ranking by pre-selecting highly-relevant candidates from the initial huge candidate pool.
	pushservice-heavy-ranker	Multi-task learning model to predict the probabilities that the target users will open and engage with the sent notifications.

Build and test code

We include Bazel BUILD files for most components, but not a top-level BUILD or WORKSPACE file. We plan to add a more complete build and test system in the future.

Contributing

We invite the community to submit GitHub issues and pull requests for suggestions on improving the recommendation algorithm. We are working on tools to manage these suggestions and sync changes to our internal repository. Any security concerns or issues should be routed to our official bug bounty program through HackerOne. We hope to benefit from the collective intelligence and expertise of the global community in helping us identify issues and suggest improvements, ultimately leading to a better X.

Read our blog on the open source initiative here.

74K ·GitHub

The-algorithm

What is this?

This is the actual source code for X's (Twitter's) recommendation algorithm — the collection of services and jobs that build and serve feeds like the For You Timeline, Search, Explore, and Notifications.

Is this one app or a collection of separate services?

It's a collection of roughly 20 separate backend services, ML models, and shared software frameworks (e.g. tweetypie, SimClusters, navi, product-mixer) that together power recommendations, not a single installable program.

How do I build and run the full stack?

You can't spin it up as a complete system yet: most components have individual Bazel BUILD files, but there's no top-level BUILD or WORKSPACE file tying them together into a runnable whole.

What languages and frameworks power the different pieces?

Most services are Scala/JVM, the ML serving layer (navi) is written in Rust for performance, and the legacy ranking framework (twml) was built on TensorFlow v1.

Can I actually self-host the full X recommendation system from this repo?

Not out of the box — this is published for transparency and study, and the repo explicitly lacks a complete, unified build and test system.

What is the heavy-ranker and why does it matter?

It's the neural network that scores sourced candidate posts, acting as the main signal that decides which posts make it into a user's For You Timeline.

What is navi used for?

Navi is the high-performance Rust-based model-serving framework that runs the ranking and recommendation models in production.

Does this repo actually cover Search and Explore?

The architecture overview mentions Search and Explore as product surfaces, but the components currently included in the repo only cover the For You Timeline and Recommended Notifications.

Why did X open source this code?

X wants the community to review the algorithm, submit issues and pull requests, and report security concerns through its official HackerOne bug bounty program.

What are similar projects?

x-algorithm, the-algorithm-ml, Chinese-Word-Vectors, predictionio, spaCy, LLMSearchRecommender, DeepLearning, Engram

Which topics is it in?

recommendation-system, scala, social-media, knowledge-graph, embeddings, microservices, search, neural-network, data-science, agpl, artificial-intelligence, distributed-systems

Creator: twitter
Size: 8.2 MB
Stars: 74K
Updated: CommitSep 03
Age: 3y
Commits · 90 Days