Bluesky study predicts post distribution with 73% accuracy
New research predicts which custom feeds will surface new posts with 73% accuracy using Bluesky's data.
Deep Dive
A new arXiv study tackles a novel cold-start problem: predicting whether newly published Bluesky posts will appear in custom feeds. The researchers built a benchmark using 5,000 monitored feeds, 17.8 million public posts, and over 1.8 million observed post–feed return records. In their evaluation, LambdaRank posted the best results across two test folds, with a capped Recall@10 of 0.7361, NDCG@10 of 0.6127, and Hit@10 of 0.7749.
Key Points
- Researchers built a benchmark using 17.8M Bluesky posts and 5K feeds to predict custom-feed returns for new posts
- LambdaRank achieved 0.7361 Recall@10 accuracy in predicting whether posts would appear in feed Top-50 results
- The study covers 9.04% of test posts with at least one positive label, representing 602K posts across two test folds
Why It Matters
This research could revolutionize social media content distribution by improving post visibility and feed curation efficiency for platforms like Bluesky.