Skip to main content

​
Overview

Home Mixer is the main service used to construct and serve Twitter’s Home Timelines. Built on Product Mixer, Twitter’s custom Scala framework for building content feeds, it currently powers:
  • For You - Best tweets from people you follow + recommended out-of-network content
  • Following - Reverse chronological tweets from people you follow
  • Lists - Reverse chronological tweets from List members

​
Architecture

Home Mixer is built on Product Mixer, a custom Scala framework that facilitates building feeds of content through structured pipelines.

​
Pipeline Structure

Product Mixer services are structured around Pipelines that split execution into transparent and structured steps:
1

Product Pipelines

Requests first go to Product Pipelines, which select which Mixer Pipeline or Recommendation Pipeline to run for a given request.
2

Mixer/Recommendation Pipelines

Each pipeline may run multiple Candidate Pipelines to fetch candidates:
  • Mixer Pipelines combine results of multiple heterogeneous Candidate Pipelines (e.g., ads, tweets, users)
  • Recommendation Pipelines score and rank results of homogeneous Candidate Pipelines
3

Candidate Pipelines

Fetch candidates from underlying Candidate Sources and perform basic operations like filtering, decoration, and feature hydration.

​
For You Timeline

The For You recommendation algorithm involves several key stages:

​
1. Candidate Generation

Fetch tweets from various candidate sources:
  • Earlybird Search Index - Real-time tweet search index
  • User Tweet Entity Graph (UTEG) - Graph-based tweet recommendations
  • CR Mixer - Candidate generation and mixing service
  • Follow Recommendations Service - Account recommendations for tweet discovery

​
2. Feature Hydration

Fetch approximately 6,000 features needed for ranking. These features include:
  • User engagement history
  • Tweet metadata and content features
  • Author features
  • Real-time signals

​
3. Scoring and Ranking

Apply machine learning models to score and rank candidates based on predicted user engagement.

​
4. Filters and Heuristics

Author Diversity

Ensure variety in tweet authors shown to users

Content Balance

Mix in-network vs out-of-network content appropriately

Feedback Fatigue

Reduce content similar to previously dismissed tweets

Visibility Filtering

Remove blocked, muted authors/tweets and apply NSFW settings

​
5. Mixing

Integrate tweets with non-tweet content:
  • Ads
  • Who-to-follow modules
  • Prompts and engagement modules

​
6. Product Features and Serving

  • Conversation modules for replies
  • Social context (“Liked by…”, “Followed by…”)
  • Timeline navigation
  • Edited tweets support
  • Feedback options
  • Pagination and cursoring
  • Client instructions and content marshalling

​
Pipeline Configurations

​
For You Pipeline Hierarchy

ForYouProductPipelineConfig
└── ForYouScoredTweetsMixerPipelineConfig
    ├── ForYouScoredTweetsCandidatePipelineConfig
    │   └── ScoredTweetsRecommendationPipelineConfig
    │       ├── ScoredTweetsInNetworkCandidatePipelineConfig
    │       ├── ScoredTweetsTweetMixerCandidatePipelineConfig
    │       ├── ScoredTweetsUtegCandidatePipelineConfig
    │       ├── ScoredTweetsFrsCandidatePipelineConfig
    │       └── ScoredTweetsScoringPipelineConfig
    ├── ForYouConversationServiceCandidatePipelineConfig
    ├── ForYouAdsCandidatePipelineConfig
    └── ForYouWhoToFollowCandidatePipelineConfig
The ForYouConversationServiceCandidatePipelineConfig serves as a backup reverse chronological pipeline in case Scored Tweets fails.

​
Following Pipeline Hierarchy

FollowingProductPipelineConfig
└── FollowingMixerPipelineConfig
    ├── FollowingEarlybirdCandidatePipelineConfig
    ├── ConversationServiceCandidatePipelineConfig
    ├── FollowingAdsCandidatePipelineConfig
    └── FollowingWhoToFollowCandidatePipelineConfig

​
Lists Pipeline Hierarchy

ListTweetsProductPipelineConfig
└── ListTweetsMixerPipelineConfig
    ├── ListTweetsTimelineServiceCandidatePipelineConfig
    ├── ConversationServiceCandidatePipelineConfig
    └── ListTweetsAdsCandidatePipelineConfig

​
Key Components

​
Product Pipeline Config

The main entry point for the For You timeline: 
@Singleton
class ForYouProductPipelineConfig @Inject() (
  forYouMixerPipelineConfig: ForYouMixerPipelineConfig,
  forYouParamConfig: ForYouParamConfig)
    extends ProductPipelineConfig[HomeMixerRequest, ForYouQuery, urt.TimelineResponse] {

  override val identifier: ProductPipelineIdentifier = ProductPipelineIdentifier("ForYou")
  // Pipeline configuration...
}
Source: home-mixer/server/src/main/scala/com/twitter/home_mixer/product/for_you/ForYouProductPipelineConfig.scala:45

​
Performance Considerations

Home Mixer handles high throughput and must maintain low latency. Key performance optimizations include:
  • Parallel candidate fetching from multiple sources
  • Efficient feature hydration with batching
  • Caching at multiple layers
  • Circuit breakers for downstream service failures

Build docs developers (and LLMs) love

Get started for freeTalk to us