How BidirectionalFollowReplyWeightBoost Amplifies Mutual Follows in X's Home Mixer

BidirectionalFollowReplyWeightBoost adds a configurable constant to the reply weight of original posts when the viewer and author maintain a reciprocal follow relationship, directly increasing the candidate's ranking score in the home-mixer feed.

The home-mixer ranking system in the xai-org/x-algorithm repository prioritizes content from mutual connections through specific weight boost mechanisms. By modifying the scoring weights for reply and dwell metrics, the system surfaces original posts from users who share a bidirectional follow relationship with the viewer while explicitly excluding replies and retweets from this amplification.

Detecting Mutual Follow Relationships

Before any weight boost applies, the system must identify whether the viewer and post author follow each other. The BidirectionalFollowHydrator in home-mixer/candidate_hydrators/bidirectional_follow_hydrator.rs handles this detection by querying the social graph and populating the is_mutual_follow_author field on each PostCandidate.

According to the source code, the hydrator checks that the viewer follows the author and the author follows the viewer, storing the result as candidate.is_mutual_follow_author (lines 57-63). This boolean flag becomes the primary signal for downstream scoring components to apply mutual-follow preferences.

Eligibility Criteria for Weight Boosting

Not all mutual-follow content receives the boost. The bidirectional_boost_eligible helper function in home-mixer/scorers/ranking_scorer.rs (lines 44-48) enforces strict eligibility requirements:

• The candidate must have no in_reply_to_tweet_id (not a reply)
• The candidate must have no retweeted_tweet_id (not a retweet)
• The is_mutual_follow_author field must be Some(true)

This filtering ensures that BidirectionalFollowReplyWeightBoost and its companion parameter BidirectionalFollowDwellWeightBoost apply exclusively to original posts, preventing amplification of conversational or redistributed content within mutual-follow relationships.

Applying the Boost to Scoring Weights

Once eligibility is confirmed, the ScoringWeights struct modifies the base weights through two dedicated methods in home-mixer/scorers/ranking_scorer.rs.

Reply Weight Enhancement The reply_weight_for method (lines 50-55) increases the base reply weight by adding self.bidirectional_follow_reply_weight_boost when the candidate meets the mutual-follow criteria.

Dwell Weight Enhancement
Similarly, dwell_weight_for (lines 78-84) adds self.bidirectional_follow_dwell_weight_boost to the base dwell weight under the same conditions.

These boosted values feed into RankingScorer::compute_weighted_score, where higher reply and dwell contributions directly increase the candidate's weighted_score and final ranking position.

Implementation Examples

Hydrating Mutual Follow Status

The following example demonstrates how BidirectionalFollowHydrator sets the mutual-follow flag that enables downstream boosting:

use home_mixer::candidate_hydrators::bidirectional_follow_hydrator::BidirectionalFollowHydrator;
use home_mixer::models::{candidate::PostCandidate, query::ScoredPostsQuery};
use std::sync::Arc;

// Mock social graph where author (id 42) follows viewer (id 1)
let hydrator = BidirectionalFollowHydrator {
    socialgraph_client: Arc::new(MockSocialGraph { followers: vec![42] })
};

let query = ScoredPostsQuery {
    user_id: 1,
    user_features: UserFeatures { followed_user_ids: vec![42], ..Default::default() },
    ..Default::default()
};

let candidate = PostCandidate { author_id: 42, ..Default::default() };
let hydrated = hydrator.hydrate(&query, &[candidate]).await;

assert_eq!(hydrated[0].as_ref().unwrap().is_mutual_follow_author, Some(true));

The hydrator populates is_mutual_follow_author = Some(true), marking the candidate for weight boosting.

Computing Boosted Weights

This example shows how ScoringWeights applies the configured boost value to the reply weight:

use home_mixer::scorers::ranking_scorer::ScoringWeights;
use home_mixer::params::BidirectionalFollowReplyWeightBoost;

// Feature switches enable a +3.0 boost
let params = FeatureSwitches::new(vec![
    ("rust_home_mixer_bidirectional_follow_reply_weight_boost", "3.0"),
    ("rust_home_mixer_reply_weight", "1.0"),
]).unwrap().match_recipient(&RecipientBuilder::new().build());

let weights = ScoringWeights::from_params(&params);
let base_reply = params.get(ReplyWeight); // 1.0

let boosted = weights.reply_weight_for(&mutual_original_candidate);
assert!((boosted - (base_reply + 3.0)).abs() < 1e-9);

The reply_weight_for method adds the configured 3.0 boost to the base 1.0 weight for eligible mutual-follow original posts.

End-to-End Scoring Impact

The boosted weights ultimately influence the final ranking through the RankingScorer:

let scorer = RankingScorer { author_cold_start: Default::default() };
let scored = scorer.score(&query, &candidates).await;
let score_with_boost = scored[0].as_ref().unwrap().score.unwrap();

// score_with_boost reflects increased reply and dwell contributions

The final score incorporates the bidirectional follow weight boosts, elevating the candidate's position in the feed.

Summary

• BidirectionalFollowReplyWeightBoost specifically targets original posts from mutual-follow relationships, excluding replies and retweets.
• The BidirectionalFollowHydrator in home-mixer/candidate_hydrators/bidirectional_follow_hydrator.rs detects mutual follows by verifying reciprocal following relationships.
• Eligibility logic in home-mixer/scorers/ranking_scorer.rs ensures only original content (no in_reply_to_tweet_id or retweeted_tweet_id) receives the boost.
• The boost adds configurable constants to both reply and dwell weights, directly increasing the candidate's final score in the ranking computation.

Frequently Asked Questions

What is BidirectionalFollowReplyWeightBoost?

BidirectionalFollowReplyWeightBoost is a configurable ranking parameter in the x-algorithm home-mixer system that adds a constant value to the reply weight of posts when the viewer and author follow each other. This mechanism prioritizes original content from mutual connections by increasing the post's overall weighted score during the ranking phase.

Does the boost apply to replies and retweets?

No. The boost explicitly applies only to original posts. The bidirectional_boost_eligible function checks that the candidate has no in_reply_to_tweet_id and no retweeted_tweet_id before applying either the reply or dwell weight boost. Replies and retweets are filtered out to ensure the amplification targets only standalone content from mutual follows.

How does the system detect mutual follows?

The system uses the BidirectionalFollowHydrator to query the social graph and verify that the viewer follows the author and the author follows the viewer. This hydrator sets the is_mutual_follow_author field on PostCandidate objects, which downstream scorers check before applying any weight boosts.

Where is the boost logic implemented?

The core boost logic resides in home-mixer/scorers/ranking_scorer.rs, specifically within the ScoringWeights implementation. The reply_weight_for and dwell_weight_for methods handle the addition of boost values, while the bidirectional_boost_eligible helper enforces content type restrictions. The mutual-follow detection occurs in home-mixer/candidate_hydrators/bidirectional_follow_hydrator.rs.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}