Abstract

Feed ranking algorithms select and prioritize what users see from a vast inventory of content on social media. They greatly impact people’s opinions, moods, and actions. Typically trained to maximize user engagement (e.g., likes, replies, and reposts), feed ranking algorithms are often blamed for exacerbating negative societal outcomes, like political polarization and toxic speech online. Until recently, running feed ranking experiments and studying the effects of feed ranking algorithms was only possible from inside social media companies. However, the emergence of middleware-based feed reranking infrastructure and more customizable platforms like Bluesky have created new opportunities for experimentation. This tutorial aims to introduce participants to these new experimental opportunities and provide a practical guide for conducting feed-ranking experiments through a mix of lectures, a case study, and hands-on exercises. The tutorial will consist of four parts: (1) an introduction to feed ranking algorithms, (2) the design of feed ranking experiments, (3) the analysis of these experiments, and (4) hands-on exercises for constructing social media feeds.

Duration: 4 hours

📅 Day: Monday, June 23, 2025
🕜 Time: 1:00 PM - 5:00 PM
📍 Location: Room 2.1.083 (AAU)

Schedule

1. History & foundations (45 mins)
2. Feed experiments using middlewares (45 mins)
3. Planning & analyzing experiments (45 mins)
4. Hands-on exercise: Build your own BlueSky feed (1 hour)

Prerequisites

• Familiarity with Python and HTTP APIs.
• We recommend having a Bluesky account.

Resources

Slides: History & foundations, Feed experiments using middlewares, Planning & analyzing experiments
Zoom meeting: Link
Hands-on execises: https://github.com/socialmediafeed25/ICWSM25BlueskyTutorial