Twitter's Recommendation Algorithm

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

Architecture

Product surfaces at Twitter 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	unified-user-actions	Real-time stream of user actions on Twitter.
	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 Tweets.
	trust-and-safety-models	Models for detecting NSFW or abusive content.
	real-graph	Model to predict the likelihood of a Twitter User interacting with another User.
	tweepcred	Page-Rank algorithm for calculating Twitter 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 Tweets from User B).
	topic-social-proof	Identifies topics related to individual Tweets.
	representation-scorer	Compute scores between pairs of entities (Users, Tweets, 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.
	Tweetypie	Core Tweet service that handles the reading and writing of Tweet data.

The product surface currently included in this repository is the For You Timeline.

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 Tweets. ~50% of Tweets come from this candidate source.
	cr-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 Tweet 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 Tweets from those accounts.
Ranking	light-ranker	Light Ranker model used by search index (Earlybird) to rank Tweets.
	heavy-ranker	Neural network for ranking candidate tweets. One of the main signals used to select timeline Tweets post candidate sourcing.
Tweet mixing & filtering	home-mixer	Main service used to construct and serve the Home Timeline. Built on product-mixer.
	visibility-filters	Responsible for filtering Twitter 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 tweets from the Earlybird Search Index and UTEG service.

Recommended Notifications

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

Type	Component	Description
Service	pushservice	Main recommendation service at Twitter used to surface recommendations to our users via notifications.
Ranking	pushservice-light-ranker	Light Ranker model used by pushservice to rank Tweets. 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 Twitter.