Skip to main content

​
Overview

CR-Mixer is a candidate generation service that speeds up iteration and development of candidate generation and light ranking. It acts as a lightweight coordinating layer that delegates candidate generation tasks to underlying compute services. The service provides:
  • Centralized platform for fetching and mixing candidate sources
  • Light ranking layer for candidates
  • Common filtering and deduplication
  • Feature switch management and version control
  • Performance optimization and caching

​
Service Definition

CR Mixer is defined in cr-mixer/thrift/src/main/thrift/cr_mixer.thrift 
service CrMixer {
  CrMixerTweetResponse getTweetRecommendations(1: CrMixerTweetRequest request)
  RelatedTweetResponse getRelatedTweetsForQueryTweet(1: RelatedTweetRequest request)
  RelatedTweetResponse getRelatedTweetsForQueryAuthor(1: RelatedTweetRequest request)
  UtegTweetResponse getUtegTweetRecommendations(1: UtegTweetRequest request)
  FrsTweetResponse getFrsBasedTweetRecommendations(1: FrsTweetRequest request)
  RelatedVideoTweetResponse getRelatedVideoTweetsForQueryTweet(1: RelatedVideoTweetRequest request)
  AdsResponse getAdsRecommendations(1: AdsRequest request)
  TopicTweetResponse getTopicTweetRecommendations(1: TopicTweetRequest request)
}

​
getTweetRecommendations

Returns personalized tweet recommendations for a user.

​
Request

​
clientContext
ClientContext
required
Client context containing user identification and device information
​
userId
int64
User ID requesting recommendations (optional for logged-out users)
​
guestId
int64
Guest ID for logged-out users
​
appId
int64
Application identifier
​
countryCode
string
User’s country code
​
languageCode
string
User’s language preference
​
product
Product
required
Product surface requesting recommendations. Enum values:
  • Home (1) - Home timeline
  • Notifications (2) - Notification recommendations
  • Email (3) - Email recommendations
  • MoreTweetsModule (4) - Related tweets module
  • ImmersiveMediaViewer (5) - Media viewer
  • VideoCarousel (6) - Video recommendations
  • ExploreTopics (7) - Topic exploration
  • Ads (8) - Ads recommendations
  • HomeRealTime (9) - Real-time home tab
  • TopicLandingPage (10) - Topic pages
​
productContext
ProductContext
Product-specific parameters and configuration options
​
excludedTweetIds
list<int64>
Tweet IDs to exclude from recommendations (already seen or dismissed tweets)

​
Response

​
tweets
list<TweetRecommendation>
List of recommended tweets with scores and metadata
​
tweetId
int64
required
Unique identifier for the recommended tweet
​
score
double
required
Recommendation score (higher is better)
​
metricTags
list<MetricTag>
Tags for tracking and attribution
​
authorId
int64
User ID of the tweet author
​
candidateGenerationKey
CandidateGenerationKey
Information about how the candidate was generated (source algorithm, similarity engine)
​
latestSourceSignalTimestampInMillis
int64
Timestamp of the most recent engagement signal used (null if not from engagement signals)

​
Exceptions

​
validationErrors
ValidationExceptionList
Validation errors with details reported to clients
​
serverError
ServerError
Server errors (details not reported to clients)

​
getRelatedTweetsForQueryTweet

Returns tweets related to a specific query tweet.

​
Request

​
internalId
InternalId
required
Internal identifier (tweet ID or user ID) to find related content for
​
product
Product
required
Product surface making the request
​
clientContext
ClientContext
required
Client context (userId may be None for logged-out users)
​
excludedTweetIds
list<int64>
Tweet IDs to exclude from results

​
Response

​
tweets
list<RelatedTweet>
List of related tweets
​
tweetId
int64
required
Tweet identifier
​
score
double
Similarity/relevance score
​
authorId
int64
Tweet author’s user ID

​
getRelatedTweetsForQueryAuthor

Returns tweets related to a specific author’s content. Uses the same request/response structure as getRelatedTweetsForQueryTweet.

​
getUtegTweetRecommendations

Returns tweet recommendations based on User Tweet Entity Graph (UTEG) engagement signals.

​
Request

​
clientContext
ClientContext
required
Client context
​
product
Product
required
Product surface
​
productContext
ProductContext
Product-specific parameters
​
excludedTweetIds
list<int64>
Tweets to exclude

​
Response

​
tweets
list<UtegTweet>
UTEG-based tweet recommendations
​
tweetId
int64
required
Tweet identifier
​
score
double
required
Sum of weights of seed users who engaged with the tweet. If a user engaged multiple times (e.g., liked and retweeted), their weight is counted for each engagement.
​
socialProofByType
map<SocialProofType, list<int64>>
required
User social proofs grouped by engagement type (likes, retweets, replies, etc.)

​
getTopicTweetRecommendations

Returns tweet recommendations for specific topics.

​
Request

​
clientContext
ClientContext
required
Client context
​
product
Product
required
Product surface
​
topicIds
list<int64>
required
List of topic IDs to get recommendations for
​
productContext
ProductContext
Product-specific parameters
​
excludedTweetIds
list<int64>
Tweets to exclude

​
Response

​
tweets
map<int64, list<TopicTweet>>
required
Map from topic ID to list of recommended tweets for that topic
​
tweetId
int64
required
Tweet identifier
​
score
double
required
Recommendation score
​
similarityEngineType
SimilarityEngineType
required
Algorithm used to generate this recommendation (e.g., SimClustersANN, TwHINANN, etc.)

​
getFrsBasedTweetRecommendations

Returns tweet recommendations based on Follow Recommendations Service (FRS) suggestions - tweets from accounts the user may be interested in following. See Thrift Definitions for complete type definitions.

Build docs developers (and LLMs) love

Get started for freeTalk to us